IL308506A - System and method for determining a score for a driver of a vehicle as part of a fleet of vehicles - Google Patents

Description

A SYSTEM AND METHOD FOR DETERMINING A SCORE FOR A DRIVER OF A VEHICLE OF A FLEET OF VEHICLES TECHNICAL FIELD The invention relates to a system and method for determining a score for a driver of a vehicle of a fleet of vehicles.

BACKGROUND Fleet vehicles are groups of motor vehicles comprising all kinds of transport vehicles that are owned by a company, a government, an agency or other business. An example of a fleet can be a public transportation company running multiple buses and other public transport vehicles of various sizes. Another example can be of a rental company managing vehicles of different types and more. Fleet owners and/or managers monitor their fleets by defining Fleet Business Metrics (FBMs). An FBM is a descriptive statistic, indicator, or figure of merit used to describe or measure something quantitatively or qualitatively that is associated with one or more vehicles of the fleets and/or with one or more drivers of vehicles of the fleet and/or with one or more passengers of vehicles of the fleet. The FBMs can include metrics to measure the state of the vehicle and/or metrics associated with the behavior of a driver while driving one or more vehicles of the fleet. The FBMs can include, for example: rate of component wear (e.g., brakes, tires, etc.), energy efficiency (e.g., fuel or electricity per unit of work, Miles Per Gallon (MPG), Miles Per Gallon of gasoline-Equivalent (MPGe), etc.), safety indicators (e.g., accident or near accident events, etc.), passenger comfort (e.g., for public transportation fleets, motion-related sickness, etc.). These FBMs can be measured for a single vehicle and/or for a group of vehicles of the fleets and/or for a single driver in one or more of his drives and/or for a group of drivers in one or more of their drives. There are many factors affecting the calculation of these FBMs. These can include contextual and environmental factors: weather, traffic, road, vehicle type and vehicle condition, vehicle load during a given drive, bus route, etc. Driver decisions (constrained by context and environment) have a significant effect on the FBMs while the effect of individual behavioral aspects might vary depending on a specific FBM.

Fleet owners and/or managers have the need to establish the impact of individual driver behavior or group of drivers (e.g., drivers belonging to a specific branch, etc.) on the FBMs. Currently fleet owners and/or managers determine the FBMs in a manual fashion that is estimated and does not rely on sensor data taken from the vehicle itself. There is thus a need in the art for a new system and method for determining a score for a driver of a vehicle of a fleet of vehicles, and specifically establish the impact of individual driver behavior or group of drivers on the FBMs.

GENERAL DESCRIPTION In accordance with a first aspect of the presently disclosed subject matter, there is provided a system for determining a score for a driver of a vehicle of a fleet of vehicles, the score is determined with respect to one or more Fleet Business Metrics (FBMs) of the fleet of vehicles, the system comprising a processing circuitry configured to: obtain: (A) a machine learning model capable of receiving an Event Density Vector (EDV) indicative of a density of unusual driver events occurring during a timeframe in which the driver is driving the vehicle and determining one or more values of the FBMs, wherein the machine learning model is trained utilizing a labeled training-data set comprising of a plurality of training records, each training record comprising: (i) a training EDV, indicative of the density of unusual driver events occurring during a given training drive on a training vehicle, and (ii) one or more training values of the FBMs corresponding to the given training drive, wherein at least one of the training values of the FBMs are calculated based on sensor readings captured from the training vehicle, and (B) a given EDV of a given driver driving the vehicle during a given timeframe; determine one or more given values of the FBMs utilizing the machine learning model and the given EDV; and calculate the score for the given driver by comparing the given values of the FBMs to other values of FBMs of other drivers from the training-data set. In some cases, the unusual driver events are driver events occurring during driving that are associated with one or more driving values that are above one or more maximal driving thresholds or below one or more minimal driving thresholds. In some cases, the maximal driving thresholds and the minimal driving thresholds are dynamically calculated for at least part of the timeframe in which the driver is driving 30 the vehicle, in accordance with one or more variables associated with driving the vehicle during the part of the timeframe. In some cases, the variables associated with driving the vehicle during the part of the timeframe include one or more of: time of day, vehicle load, road geometry, and movement trajectory of the vehicle over the part of the timeframe. In some cases, the dynamic calculations are based on a second machine learning model, capable of receiving one or more variables describing a road segment that is traveled during the part of the timeframe by the driver driving the vehicle and determining the corresponding maximal driving thresholds and the corresponding minimal driving thresholds. In some cases, the driver events include one or more of: longitude velocity, latitude velocity, longitude acceleration, latitude acceleration, longitude jerk, and latitude jerk. In some cases, the given EDV is calculated based on the density of unusual driver events occurring during the given timeframe. In some cases, the given EDV is normalized over a distance driven during the given timeframe. In some cases, the processing circuitry is further configured to determine an explanation of the score calculated for the given driver, wherein the explanation is based on weights of the machine learning model. In some cases, the machine learning model is a linear Non-Negative Least Squares (NNLS) regression model. In accordance with a second aspect of the presently disclosed subject matter, there is provided a system for determining one or more driving thresholds for a road segment, the system comprising a processing circuitry configured to: obtain: (A) a machine learning model capable of receiving: (i) one or more road segment variables describing the road segment that is traveled during a timeframe, and (ii) one or more driving conditions variables describing driving conditions associated with an area of the road segment during the timeframe, and determining the driving thresholds for the given road segment, (B) one or more given road segment variables of a given road segment that is traveled during a given timeframe, and (C) one or more given driving conditions variables associated with a given area of the given road segment during the given timeframe; and determine the driving thresholds for the given road segment by utilizing the machine learning model and the given road segment variables. In accordance with a third aspect of the presently disclosed subject matter, there is provided a method for determining a score for a driver of a vehicle of a fleet of vehicles, the score is determined with respect to one or more Fleet Business Metrics (FBMs) of the fleet of vehicles, the method comprising: obtain, using a processing circuitry: (A) a machine learning model capable of receiving an Event Density Vector (EDV) indicative of a density of unusual driver events occurring during a timeframe in which the driver is driving the vehicle and determining one or more values of the FBMs, wherein the machine learning model is trained utilizing a labeled training-data set comprising of a plurality of training records, each training record comprising: (i) a training EDV, indicative of the density of unusual driver events occurring during a given training drive on a training vehicle, and (ii) one or more training values of the FBMs corresponding to the given training drive, wherein at least one of the training values of the FBMs are calculated based on sensor readings captured from the training vehicle, and (B) a given EDV of a given driver driving the vehicle during a given timeframe; determine, using the processing circuitry, one or more given values of the FBMs utilizing the machine learning model and the given EDV; and calculate, using the processing circuitry, the score for the given driver by comparing the given values of the FBMs to other values of FBMs of other drivers from the training-data set. In some cases, the unusual driver events are driver events occurring during driving that are associated with one or more driving values that are above one or more maximal driving thresholds or below one or more minimal driving thresholds. In some cases, the maximal driving thresholds and the minimal driving thresholds are dynamically calculated for at least part of the timeframe in which the driver is driving the vehicle, in accordance with one or more variables associated with driving the vehicle during the part of the timeframe. In some cases, the variables associated with driving the vehicle during the part of the timeframe include one or more of: time of day, vehicle load, road geometry, and movement trajectory of the vehicle over the part of the timeframe. In some cases, the dynamic calculations are based on a second machine learning model, capable of receiving one or more variables describing a road segment that is traveled during the part of the timeframe by the driver driving the vehicle and determining the corresponding maximal driving thresholds and the corresponding minimal driving thresholds. In some cases, the driver events include one or more of: longitude velocity, latitude velocity, longitude acceleration, latitude acceleration, longitude jerk, and latitude jerk. In some cases, the given EDV is calculated based on the density of unusual driver events occurring during the given timeframe. In some cases, the given EDV is normalized over a distance driven during the given timeframe. In some cases, the method is further configured to determine, using the processing circuitry, an explanation of the score calculated for the given driver, wherein the explanation is based on weights of the machine learning model. In some cases, the machine learning model is a linear Non-Negative Least Squares (NNLS) regression model. In accordance with a fourth aspect of the presently disclosed subject matter, there is provided a method for determining one or more driving thresholds for a road segment, the method comprising: obtain, using a processing circuitry: (A) a machine learning model capable of receiving: (i) one or more road segment variables describing the road segment that is traveled during a timeframe, and (ii) one or more driving conditions variables describing driving conditions associated with an area of the road segment during the timeframe, and determining the driving thresholds for the given road segment, (B) one or more given road segment variables of a given road segment that is traveled during a given timeframe, and (C) one or more given driving conditions variables associated with a given area of the given road segment during the given timeframe; and determine, using the processing circuitry, the driving thresholds for the given road segment by utilizing the machine learning model and the given road segment variables. In accordance with a fifth aspect of the presently disclosed subject matter, there is provided a non-transitory computer readable storage medium having computer readable program code embodied therewith, the computer readable program code, executable by processing circuitry of a computer to perform a method comprising: obtain, using a processing circuitry: (A) a machine learning model capable of receiving an Event Density Vector (EDV) indicative of a density of unusual driver events occurring during a timeframe in which the driver is driving the vehicle and determining one or more values of the FBMs, wherein the machine learning model is trained utilizing a labeled training-data set comprising of a plurality of training records, each training record comprising: (i) a training EDV, indicative of the density of unusual driver events occurring during a given training drive on a training vehicle, and (ii) one or more training values of the FBMs corresponding to the given training drive, wherein at least one of the training values of the FBMs are calculated based on sensor readings captured from the training vehicle, and (B) a given EDV of a given driver driving the vehicle during a given timeframe; determine, using the processing circuitry, one or more given values of the FBMs utilizing the machine learning model and the given EDV; and calculate, using the processing circuitry, the score for the given driver by comparing the given values of the FBMs to other values of FBMs of other drivers from the training-data set. In accordance with a sixth aspect of the presently disclosed subject matter, there is provided a non-transitory computer readable storage medium having computer readable program code embodied therewith, the computer readable program code, executable by processing circuitry of a computer to perform a method comprising: obtain, using a processing circuitry: (A) a machine learning model capable of receiving: (i) one or more road segment variables describing the road segment that is traveled during a timeframe, and (ii) one or more driving conditions variables describing driving conditions associated with an area of the road segment during the timeframe, and determining the driving thresholds for the given road segment, (B) one or more given road segment variables of a given road segment that is traveled during a given timeframe, and (C) one or more given driving conditions variables associated with a given area of the given road segment during the given timeframe; and determine, using the processing circuitry, the driving thresholds for the given road segment by utilizing the machine learning model and the given road segment variables. BRIEF DESCRIPTION OF THE DRAWINGSIn order to understand the presently disclosed subject matter and to see how it may be carried out in practice, the subject matter will now be described, by way of non-limiting examples only, with reference to the accompanying drawings, in which: Fig. 1 is a schematic illustration of an exemplary segment trajectory box, in accordance with the presently disclosed subject matter; Fig. 2 is a block diagram schematically illustrating one example of a system for driver score determination for a driver of a vehicle of a fleet of vehicles, in accordance with the presently disclosed subject matter; Fig. 3 is a flowchart illustrating one example of a sequence of operations carried out for score determination for a driver of a vehicle of a fleet of vehicles, in accordance with the presently disclosed subject matter; and Fig. 4 is a flowchart illustrating one example of a sequence of operations carried out for driving thresholds determination for a road segment, in accordance with the presently disclosed subject matter.

DETAILED DESCRIPTION In the following detailed description, numerous specific details are set forth in order to provide a thorough understanding of the presently disclosed subject matter. However, it will be understood by those skilled in the art that the presently disclosed subject matter may be practiced without these specific details. In other instances, well-known methods, procedures, and components have not been described in detail so as not to obscure the presently disclosed subject matter. In the drawings and descriptions set forth, identical reference numerals indicate those components that are common to different embodiments or configurations. Unless specifically stated otherwise, as apparent from the following discussions, it is appreciated that throughout the specification discussions utilizing terms such as "generating", "obtaining", "training", "identifying", "calculating", "providing", "executing" or the like, include action and/or processes of a computer that manipulate and/or transform data into other data, said data represented as physical quantities, e.g., such as electronic quantities, and/or said data representing the physical objects. The terms "computer", "processor", "processing resource", "processing circuitry" and "controller" should be expansively construed to cover any kind of electronic device with data processing capabilities, including, by way of non-limiting example, a personal desktop/laptop computer, a server, a computing system, a communication device, a smartphone, a tablet computer, a smart television, a processor (e.g. digital signal processor (DSP), a microcontroller, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), etc.), a group of multiple physical machines sharing performance of various tasks, virtual servers co-residing on a single physical machine, any other electronic computing device, and/or any combination thereof. The operations in accordance with the teachings herein may be performed by a computer specially constructed for the desired purposes or by a general-purpose computer specially configured for the desired purpose by a computer program stored in a non-transitory computer readable storage medium. The term "non-transitory" is used herein to exclude transitory, propagating signals, but to otherwise include any volatile or non-volatile computer memory technology suitable to the application. As used herein, the phrase "for example," "such as", "for instance" and variants thereof describe non-limiting embodiments of the presently disclosed subject matter. Reference in the specification to "one case", "some cases", "other cases" or variants thereof means that a particular feature, structure or characteristic described in connection with the embodiment(s) is included in at least one embodiment of the presently disclosed subject matter. Thus, the appearance of the phrase "one case", "some cases", "other cases" or variants thereof does not necessarily refer to the same embodiment(s). It is appreciated that, unless specifically stated otherwise, certain features of the presently disclosed subject matter, which are, for clarity, described in the context of separate embodiments, may also be provided in combination in a single embodiment. Conversely, various features of the presently disclosed subject matter, which are, for brevity, described in the context of a single embodiment, may also be provided separately or in any suitable sub-combination. In embodiments of the presently disclosed subject matter, fewer, more and/or different stages than those shown in Figs . 3-4 may be executed. In embodiments of the presently disclosed subject matter one or more stages illustrated in Figs. 3-4 may be executed in a different order and/or one or more groups of stages may be executed simultaneously. Figs . 1-2 illustrate a general schematic of the system architecture in accordance with an embodiment of the presently disclosed subject matter. Each module in Figs . 1-2can be made up of any combination of software, hardware and/or firmware that performs the functions as defined and explained herein. The modules in Figs . 1-2may be centralized in one location or dispersed over more than one location. In other embodiments of the presently disclosed subject matter, the system may comprise fewer, more, and/or different modules than those shown in Figs. 1-2 . Any reference in the specification to a method should be applied mutatis mutandis to a system capable of executing the method and should be applied mutatis mutandis to a non-transitory computer readable medium that stores instructions that once executed by a computer result in the execution of the method. Any reference in the specification to a system should be applied mutatis mutandis to a method that may be executed by the system and should be applied mutatis mutandis to a non-transitory computer readable medium that stores instructions that may be executed by the system. Any reference in the specification to a non-transitory computer readable medium should be applied mutatis mutandis to a system capable of executing the instructions stored in the non-transitory computer readable medium and should be applied mutatis mutandis to method that may be executed by a computer that reads the instructions stored in the non-transitory computer readable medium. The described driver scoring system allows fleet owners and/or managers to automatically establish the impact of the behavior of an individual driver or group of drivers (e.g., drivers belonging to a specific branch, etc.) on the FBMs. The system scores drivers, groups of drivers, or any combination of drivers and/or vehicle types and/or route properties and/or number and/or passengers of the vehicle, etc. according to their impact on at least one FBM during a given time interval. The described system associates specific driver events (for example: harsh braking, over and/or under speeding, excessive acceleration, etc.) with an increased and/or decreased driver score. The described system can be used to adjust incentives for drivers, as driver compensation can depend on the driver's performance as determined by the system. The system can be utilized to educate drivers: explain and help drivers to improve their driver scores. The described system can also be used to provide near real-time feedback to the driver. For example: by utilizing an in-cabin indicator indicative of the changing driver score in near real-time. The in-cabin indicator can also be indicative of impactful driver events. The described system can also provide periodic reports relating to driver behavior analytics. Such reports can include for example, for a given driver: (i) Safety elements for that driver, such as: FBM scores, impact on the FBM scores by event types, rank of the driver in comparison to other drivers of vehicles in the fleet for these safety elements, (ii) Wear for a given component of a vehicle driven by that driver, such as: FBM scores, impact on the FBM scores by event types, rank of the driver in comparison to other drivers of vehicles in the fleet for these wear elements, (iii) Passenger comfort elements for passengers driven by that driver, such as: FBMs scores, impact on the FBMs score by event types, rank of the driver in comparison to other drivers of vehicles in the fleet for these passenger comfort elements, and (iv) Efficiency elements for that driver, such as: FBMs scores, impact on the FBMs score by event types, rank of the driver in comparison to other drivers of vehicles in the fleet for these efficiency elements. The system can provide drill-down and slice-n-dice dashboard and/or analytics as part of the report to show individual events and/or to aggregate the information in accordance with time and/or location and/or event type and/or other parameters. The described system can perform the following steps on order to determine the driver score and to automatically establish the impact of the behavior of an individual driver or group of drivers (e.g., drivers belonging to a specific branch, etc.) on the FBMs defined by the fleet managers: - Detect unusual driver events given context and environment variables. There are multiple event types: extreme positive and/or negative acceleration and/or jerk and/or velocity in longitudinal and/or latitudinal directions. The output of this step is a ride Event Counter Vector (ECV). The size of the ECV corresponds to the number of different event types. - Normalize the ECV and produce unusual Event Density Vector (EDV) for each ride by aggregating and normalizing the ECV per driving duration (which can exclude idle time) and/or per any other measurement that can be used to normalize the ECV. The output of this step is a ride EDV of the same size as the ECV. - Prepare Fleet Business Metrics (FBMs) values for each ride – these are computed as an explicit FBM (e.g., using measured MPG, or any other measurement associated with the vehicle and/or the driver of the vehicle and/or passengers of the vehicle) or implicit FBM proxy values (e.g., ratio of brake pedal to deacceleration as an indicator for brake wear, or any other implicit indicator of a measured value). The output of this step is a ride FBMs values vector of a size equal to the number of FBMs. - Train a regression model to predict a score from normalized event counters (EDV), for example, by using a Non-Negative Least Square (NNLS) regression model. The output of this step is a prediction model per FBM from ride EDV to ride FBM. - Score and explain any collection of rides using the trained model. Compute driver scores and explain the impact of each type of event for a specific driver and/or group of drivers. The output of this step are scores for each driver and/or group of drivers and impact weights of each type of event for each score and/or for each driver. These steps are further detailed herein, inter alia with reference to Fig. 1. It is to be noted that at least some of the values measured for the properties of the vehicle's trajectory and/or movement are measured using on-board sensors. Such as: Global Positioning System (GPS) sensor, Inertial Measurement Unit (IMU), accelerometer, g-sensor, gyroscope sensor, wheel velocity sensor, or any other on-board and/or off-board sensor associate with the vehicle and/or the driver of the vehicle and/or the passenger of the vehicle. These measurements can include velocity/acceleration/jerk (wherein jerk is the rate of change of the acceleration). The measurements are measured in a consistent coordinate system (for example: in the vehicle's internal coordinate system). The measurements can undergo calibration and filtering. Bearing this in mind attention is drawn to Fig. 1 , which is a schematic illustration of an exemplary segment trajectory box, in accordance with the presently disclosed subject matter. A driver event (e.g., driver event A 120-a, driver event B 120-b, driver event C 120-c, …, driver event N 120-n) is a measurement during a given timeframe of one or more parameters associated with a drive (or a ride) in which an individual driver or a group of drivers are driving one or more vehicles that are part of a fleet of vehicles. A non-limiting example is of measuring the velocity and/or acceleration and/or jerk of a given vehicle at a certain time during a drive made by the driver. The measurements can be made over two axes: longitude and latitude. These values measured make up the driver event (e.g., driver event A 120-a, driver event B 120-b, driver event C 120-c, …, driver event N 120-n). The measurements can be of the maximum and minimum values measured for the parameters over a timeframe. In a non-limiting example, each driver event (e.g., driver event A 120-a, driver event B 120-b, driver event C 120-c, …, driver event N 120-n) can be a vector of twelve measured values – three measurements (velocity, acceleration, jerk) in two axes (longitude, latitude) in two extremums (minimum, maximum) – as measured over a timeframe of the drive. In this non-limiting example, a given driver event (e.g., driver event A 120-a, driver event B 120-b, driver event C 120-c, …, driver event N 120-n) can be represented by one or more measured values, measured in two axes (longitude, latitude), which can include for example: a maximal longitude and/or latitude velocity of the given vehicle driven by the driver during a timeframe, a minimal longitude and/or latitude velocity of the given vehicle driven by the driver during the timeframe, a maximal longitude and/or latitude acceleration of the given vehicle driven by the driver during the timeframe, a minimal longitude and/or latitude acceleration of the given vehicle driven by the driver during the timeframe, a maximal longitude and/or latitude jerk of the given vehicle driven by the driver during the timeframe, a minimal longitude and/or latitude jerk of the given vehicle driven by the driver during the timeframe. These measured values can be depicted as a vector representative of the given driver event. A Segment Trajectory Box (STB) 110 is defined as the admissible values of the driver event parameters for a given road segment traveled by the driver driving the vehicle during the drive under certain driving and environment conditions. The STB 110 can be for example an area in six-dimensional space – having three measurements, two axis and two extremums. A given driver event (e.g., driver event A 120-a, driver event B 120-b, driver event C 120-c, …, driver event N 120-n) can be depicted as a point in this six-dimensional space. A given exemplary driver event in accordance with the measured values is within the STB 110 is an admissible driver event. Another exemplary driver event that has measured values that are outside of the STB 110, with respect to at least one of the dimensions of the STB 110, is an unusual driver event. These unusual driver events are used in the following phases detailed below by system 200 as the basis of scoring a driver. These unusual driver events are used to compute an Event Counter Vector (ECV) – a vector of counters, each counting the number of unusual driver events occurring during a segment of a drive, The STB 110 can be dynamically determined by the system, for example: an STB 110 is determined for every segment of the road driven during the drive. A segment of the road has a certain geometry, and a driver needs to decide on a specific trajectory for passing the road segment given context and/or environment variables. The context variable can include, for example: time of day, vehicle load, etc. The segment's geometry can be for example: a left turn of a specific curvature. The trajectory taken by the driver for that segment under those conditions can be described as movement along the geometry as a function of time. A trajectory can be described by its extremum values – maximum and minimum of velocity, acceleration and jerk over two axes of longitude and latitude. As mentioned above, the dynamically determined STB 110 can be used to determine if a driver event (e.g., driver event A 120-a, driver event B 120-b, driver event C 120-c, …, driver event N 120-n) is an admissible driver event with a usual trajectory by looking at the driver events (e.g., driver event A 120-a, driver event B 120-b, driver event C 120-c, …, driver event N 120-n) as an input defined by twelve values is within the STB 110 in six-dimensional space. Unusual driver events occur when a specific trajectory is out of admissible six-dimensional STB 110. Optionally and/or additionally, the system can measure the magnitude of the driver event – how far is that given driver event from the STB 110 and/or the duration in which the given driver event is out of the STB 110. In some cases, the described system can determine the STB 110 by utilizing a machine learning model capable of predicting a corresponding STB 110 for a given road segment. The machine learning model can be trained using historical data. A non-limiting example of such a machine learning model is a regression model trained to predict STB 110 from a segment's contextual variable and geometry representation. An average or usual STB 110 can be determined based on one or more historical drives by drivers passing through this segment and/or similar segments under similar conditions. In some cases, the geometry representation can be normalized by the system so that similar segment geometry (for example: straight segments or sharp left-turn segments) will have similar representation. In these cases, it can be assumed that similar geometry representations having similar contextual variables will produce similar STBs 110. The machine learning model can be an optimized regression model, optimized for minimizing average Intersection over Union (IoU) metrics in six-dimensional space. For example, by using a Jaccard similarity coefficient between the predicted and the actual STBs 110. Optionally, a generalized IoU evaluation metric can be used for the regression model. A non-limiting example of utilizing segment geometry representations for the training of the regression model can be segmenting (optionally, with an overlap) the entire drive into one or more segments of fixed length (e.g., 50m segments). The segmentation can be performed by interpolating fixed rate GPS samples (e.g., at a 1Hz sampling rate). In this non-limiting example, the system generates the training data by re-encoding the segments using a fixed length coordinate vector (e.g., a ten coordinate (x, y) pairs vector), interpolating the segments for compensating the difference in speed (for example: a fast-moving vehicle can pass 50m with only a single GPS sample in 1 sec), re-encoding the segments using relative coordinates by subtracting a first point coordinate from other points so each segment starts at a common origin. The segments can be rotated for a normalized direction by using Principal Components Analysis (PCA) or a similar method and rotate the segment geometry to align the primary direction with the Y axis. An exemplary regression model training dataset consists of training records, wherein a training record of the training records comprises of: metadata fields (e.g., segment id, driver id, etc.), context variables (e.g., time of day, vehicle attributes, etc.), context proxy variables (e.g., average speed, approximate location, etc.), normalized road segment geometry (e.g., segment geometry representation, zero started, rotated, fixed length, etc.), ground truth vector (e.g., six-dimensional actual STB 110, vector of size twelve, etc.). In some cases, the training data comprises context proxy variables. These are used because not all contextual variables are directly observable and/or known (for example: state of the traffic light). As unobservable variables can greatly affect the trajectory (for example: do/don’t do a traffic light stop), the system can determine one or more proxy variable in the training of the machine learning model. Examples can include: average speed proxy variable as a proxy for the driving mode (e.g., traffic light, traffic jam, etc.), rough GPS location variable as a proxy for road type (e.g., in-city, highway, etc.). The system can utilize the machine learning model for computing a drive (or ride) ECV – a vector of counters, each counting the number of unusual driver events occurring during a segment of a drive. The ECV can be computed for at least one of the drives in which an individual driver or a group of drivers are driving one or more vehicles that are part of the fleet of vehicles. The system can compare the actual trajectory of at least one segment of the segments of the given drive with the segment’s STB 110 prediction given by the machine learning model. The predicted six-dimensional STB 110 can be defined by twelve "threshold" values (minimum and maximum along each axis). For each segment the system can construct a twelve-dimensional binary vector. The system can set values of the vector to one if the corresponding threshold was crossed in this segment. In addition, the system can use a non-binary vector with rich data for counting the number of drivee events outside the predicted STB 110. The system can take into account the magnitude of each driver event outside of the predicted STB 110 (e.g., how far from the threshold) and/or the duration of the driver event being outside the predicted STB 1(e.g., how long the trajectory was outside the STB 110). The system can then compute the ECV by summing up the determined binary vectors for all segments in the given drive. Optionally, the system can compensate for double counting due to at least partial segment overlap. After computing the ECV, the system can determine an Event Density Vector (EDV) being a normalization of the ECV. The ECVs can have larger values for longer rides, as there is more time for unusual driver events to be created over the longer timeframe of the long drive. The system determines a rate of unusual driver events (driver events that are outside the STB 110) for at least one drive. The rate is the density of the unusual driver events, for example, over time. The system can exclude idle time (long period of zero movement and/or velocity and/or jerk). In some cases, the system cannot directly use the drive duration as it might include long parking times. In order to determine the EDV, which is a vector of the size of the ECV, the system divides the values of the EDV by the net driving time of the corresponding drive. The system can now utilize the EDVs to determine values for the FBMs for at least one drive. In some cases, determining the FBMs is explicit, such as: an energy efficiency FBM that can be measured as MPG or MPGe. In some cases, the FBMs are not readily available for measurement – they are implicit FBMs. These implicit FBMs can be deduced by the system from other information available to the system – such as: information that exists not at the drive resolution (e.g., brakes replacement events) or that can be estimated from the data (e.g., comfort of a passenger during the drive). The system can use, for example, the following methods to deduce the implicit FBMs: Synthetic Wear Signal (SWR), Excessive Wear Rate (EWR), Partial Order Rank (POR) and other methods. In a non-limiting example, SWR is used to compute a ride wear FBM by averaging over the entire ride the given deacceleration measurements and brake pedal position signals to compute the ratio deacceleration/pedal position. The ride wear FBM should decrease when brakes are worn. Another non-limiting example is utilizing EWR to compute a ride wear FBM by measuring the excessive wear rate for a given drive: the usual replacement distance is every R kilometer of driving (this information can be retrieved from manufacturer instructions for the vehicle used for the drive). The actual replacement distance: M kilometer (this information can be retrieved from service logs of the vehicle used for the drive). The ride wear FBM can be computed as the drive distance times the ratio of M/R. The ride wear FBM increases with excessive rates. Another non-limiting example is utilizing POR to compute a ride comfort FBM by utilizing a motion sickness and/or comfort events density vector. This is performed by converting drive data into a vector of counters of physiological parameters. These physiological parameters define motion sickness and/or comfort events. There are several standards (such as: ISO 2631 – Mechanical vibration and shock — Evaluation of human exposure to whole-body vibration) and physiological experiments (such as the one described in: "BAE, Il, et al. Self-driving like a human driver instead of a robocar: Personalized comfortable driving experience for autonomous vehicles. arXiv preprint arXiv:2001.03908, 2020") that define the human body's response to motion in the context of transportation. The system can convert the drive information into a vector of counters of physiological events that are defined by experimentally obtained motion sickness and/or comfort thresholds. The system can then transform the motion sickness and/or comfort events density vector into the ride comfort FBM. This can be achieved, for example, by taking as an input a vector of size eight for a given drive: two metrics (acceleration and jerk), two axes (longitude and latitude), and two severity levels (aggressive and extremely aggressive) which is a Comfort Event Vector (CEV), that can have normalized counters per the given drive duration. The output in this non-limiting example is a comfort FBM, which is a single number reflecting overall passenger comfort for the given drive. The system can also assign different ranks to the CEV in order to decide which of the parameters has a greater influence on the comfort FBM. For example, does two latitude extreme accelerations or ten aggressive longitude jerks have more influence on the comfort FBM? The system can utilize POR to assign different ranks to the CEV. One non-limiting example of a method for using POR is: - Step 1: Obtain training dataset for comfort FBM. This training dataset can be different from the driver events training dataset. Compute CEV for each ride using physiologically defined thresholds. - Step 2: Induce partial order on the computed CEV. Define X<=Y if all X values are less than corresponding Y values within a corresponding CEV. - Step 3: Define CEV "rank". Rank of a given CEV X can be defined for example as the ratio of CEVs that are "better" than X as compared to at least one CEV in the training dataset. - Step 4: Train a regression model (for example: using a non-negative least square algorithm). Predict "rank" from CEV using resulting linear formula. - Step 5: Use resulting regression model for establishing drive comfort FBM (target). Compute CEV based on one or more pre-defined physiological thresholds. Run regression model to compute drive comfort FBM. In this exemplary method, the system can define partial order and rank of the CEVs by bucketizing each value of the CEV into one or more possible values. For example: for ten possible values, where the 0th bucket has no events of this type and the 10th bucket had a maximal number of events. The system can then define a partial order between CEV X<=Y if all elements of X are less or equal to the corresponding element of Y. Some X and Y of the CEV are incomparable as some of X elements might be greater, and some might be less than those of Y. The system can then define the rank of a given CEV X as number of Y such that Y<=X divided by the total number of elements in the training data. A CEV where almost all other CEVs of the training data are "better" thereof, will receive a higher rank value. It is to be noted that there is a difference between driver events (e.g., driver event A 120-a, driver event B 120-b, driver event C 120-c, …, driver event N 120-n) and comfort events. The Driver Event Vector is based on normal behavior of all drivers in the fleet of vehicles. The Comfort Events Vector is based on physiologically defined thresholds. Driver events are used by the system together with context variables for predicting different FBMs (such as: safety FBMs, wear FMS, etc.). Comfort events are used for defining a single comfort FBM to be used as a target of the machine learning model. Comfort events can be used by the system for scoring directly. But it is to be noted that comfort events might occur due to context variables (e.g., a traffic jam, etc.) that are beyond the driver's control while some driver decisions might lead, but not directly corresponds to motion sickness and/or comforts events (e.g., over speeding, etc.). The system can now predict the FBMs based on the EDVs. The system can train and utilize one or more models to be used for different FBMs. For example, using an NNLS regression model. Linear models are explainable regarding the resulting driver scores. These models have a prediction formula of: a1*e1+ …+a12*e12, wherein e1-eare events density (event counters normalized by net driving time) and a1-a12 are nonnegative weights determined by the model. The non-negative weights ensure that more unusual events lead to worse scores. It is to be noted that the system can calculate drive scores, which are score for a given drive. The drive can comprise one or more road segments that are driven through. A drive score can include scoring each of these road segments. The system can compute a driver score, which is a score for a driver that has driven the corresponding drives. The driver score is based on the calculated drive scores. The system can now utilize one or more models to score one or more drivers of the vehicles of the fleet and optionally explain the reason for their respective scores. Given an FBM prediction model (for example: a regression model) from driver event density vector FBM(e), the system can compute the score for a drive as Score(e)=FBM(e)-FBM(0), wherein e is EDV vector and 0 is all zero vector and/or 0 is an average of the scores for at least some of the drivers of the fleet of vehicles. The final driver score is a (weighted) average of driver’s drives scores. Optionally, the system can weigh the drive score by duration or distance of that given drive. The system can determine the impact of event of type I as a[i] x e[i], wherein e[i] is the ith element of the EDV and a[i] is the corresponding weight in the linear NNLS regression model. It is to be noted that the system can also utilize driver control events. Determining the segment trajectory by various driving control methods, such as: braking using pedal brakes or deceleration using engine and/or inertia, acceleration using different gears position and Rounds Per Minute (RPM). While the vehicle trajectory might not be affected by specific choice of the driver control strategy it might have a significant effect on some FBMs (e.g., on a wear FBM). With these additional driving control methods, the system can extend the STB with control parameters: the system assumes that the control inputs have numerical/ordeal representation (for example: gear position, brake and/or acceleration pedal position, steering wheel position). The six-dimensional STB described above can be extended into additional dimensions using the relevant control dimensions. The dimension of ground truth vector, regression output and event counter vector are increased proportionally. For example, the system can include a brake pedal position resulting in a seven-dimensional STB with a fourteen-element ECV. Having briefly described the steps that can be taken in order to determine the driver score and to automatically establish the impact of the behavior of an individual driver or group of drivers (e.g., drivers belonging to a specific branch, etc.) on the FBMs defined by the fleet managers, attention is drawn to Fig. 2 , a block diagram schematically illustrating one example of a system for driver score determination for a driver of a vehicle of a fleet of vehicles, in accordance with the presently disclosed subject matter. According to certain examples of the presently disclosed subject matter, system 200 (please note that the terms "driver scoring system" and "system" are used herein interchangeably) can comprise a network interface 220 enabling connecting the system 200 to a network and enabling it to send and receive data sent thereto through the network, including in some cases receiving information such as: one or more machine learning models, one or more Fleet Business Metrics (FBMs) and their corresponding values, one or more driver events (e.g., driver event A 120-a, driver event B 120-b, driver event C 120-c, …, driver event N 120-n), one or more Event Density Vectors (EDVs), one or more Comfort Event Vectors (CEVs), training records, driver scores, average driver scores for at least some of the drivers of the fleet, one or more regression models, etc. In some cases, the network interface 220 can be connected to a Local Area Network (LAN), to a Wide Area Network (WAN), to a Controller Area Network bus (CAN-bus), or to the Internet. In some cases, the network interface 220 can connect to a wireless network. It is to be noted that in some cases the information, or part thereof, is transmitted from a physical entity (for example: from a vehicle that is part of a fleet). System 200 can further comprise or be otherwise associated with a data repository 210 (e.g., a database, a storage system, a memory including Read Only Memory – ROM, Random Access Memory – RAM, or any other type of memory, etc.) configured to store data, including, inter alia, one or more machine learning models, one or more Fleet Business Metrics (FBMs) and their corresponding values, one or more driver events (e.g., driver event A 120-a, driver event B 120-b, driver event C 120-c, …, driver event N 120-n), one or more Event Density Vectors (EDVs), one or more Comfort Event Vectors (CEVs), training records, driver scores, average driver scores for at least some of the drivers of the fleet, one or more regression models, etc. In some cases, data repository 210 can be further configured to enable retrieval and/or update and/or deletion of the data stored thereon. It is to be noted that in some cases, data repository 210 can be distributed. It is to be noted that in some cases, data repository 210 can be stored in cloud-based storage. System 200 further comprises processing circuitry 230. Processing circuitry 230 can be one or more processing circuitry units (e.g., central processing units), microprocessors, microcontrollers (e.g., microcontroller units (MCUs)) or any other computing devices or modules, including multiple and/or parallel and/or distributed processing circuitry units, which are adapted to independently or cooperatively process data for controlling relevant system 200 resources and for enabling operations related to system 200 resources. The processing circuitry 230 comprises a driver score determination module 240, configured to perform a driver score determination process, as further detailed herein, inter alia with reference to Fig. 3. The processing circuitry 230 can further comprise a driving thresholds determination module 250, configured to perform a driving thresholds determination process, as further detailed herein, inter alia with reference to Fig. 4. Turning to Fig. 3 , a flowchart illustrating one example of a sequence of operations carried out for score determination for a driver of a vehicle of a fleet of vehicles, in accordance with the presently disclosed subject matter. According to certain examples of the presently disclosed subject matter, system 200 can be configured to perform a score determination process 300, e.g., utilizing the driver score determination module 240. System 200 can determine one or more scores for one or more drivers of vehicles belonging to a fleet of vehicles. The scores are determined in accordance with Fleet Business Metrics (FBMs). The FBMs are defined by a fleet owner and/or manager. An FBM is a descriptive statistic, indicator, or figure of merit used to describe or measure something quantitatively or qualitatively that is associated with one or more vehicles of the fleet and/or with one or more drivers of vehicles of the fleet and/or with one or more passengers of vehicles of the fleet. The FBMs can include metrics to measure the state of the vehicle and/or metrics associated with the behavior of a driver while driving one or more vehicles of the fleet. The FBMs can include, for example: rate of component wear (e.g., brakes, tires, etc.), energy efficiency (e.g., fuel or electricity per unit of work, Miles Per Gallon (MPG), Miles Per Gallon of gasoline-Equivalent (MPGe), etc.), safety indicators (e.g., accident or near accident events, etc.), passenger comfort (e.g., for public transportation fleets, motion-related sickness, etc.). These FBMs can be measured for a single vehicle and/or for a group of vehicles of the fleet and/or for a single driver in one or more of his drives and/or for a group of drivers in one or more of their drives. The system can optionally also determine an explanation of the score determined for the driver. For example, a given driver has driven 10 drives with a vehicle that is part of the fleet. The system can determine one or more scores for one or more FBMs for the given driver based on analysis of these 10 drives. In this example, system 200 determined that the score of the passenger comfort is 8. The system 200 can also determine an explanation of why the driver got a score of 8 for the passenger comfort FBM. For example: because of high rate of acceleration changes in comparison to the behavior of other drivers of the fleet. The driver score can be relative to the scores of one or more of the other drivers in the fleet, for example: in a fleet of buses, the driver score can be compared with a group of drivers driving the same route. The system can thus determine the impact of the given driver on the fleet. For this purpose, system 200 can be configured to obtain: (A) a machine learning model capable of receiving an Event Density Vector (EDV) indicative of a density of unusual driver events occurring during a timeframe in which the driver is driving the vehicle and determining one or more values of the Fleet Business Metrics (FBMs), wherein the machine learning model is trained utilizing a labeled training-data set comprising of a plurality of training records, each training record comprising: (i) a training EDV, indicative of the density of unusual driver events occurring during a given training drive on a training vehicle, and (ii) one or more training values of the FBMs corresponding to the given training drive, wherein at least one of the training values of the FBMs are calculated based on sensor readings captured from the training vehicle, and (B) a given EDV of a given driver driving the vehicle during a given timeframe (block 310). The calculation of the training values of the FBMs can be direct (for example: using a speed value by reading the speed of the vehicle) or indirect using a proxy value (for example: for a rate of component wear value a proxy value of the rate of pressing the brake pedal can be used to determine the rate of component wear for the breaks). Unusual driver events are driver events (e.g., driver event A 120-a, driver event B 120-b, driver event C 120-c, …, driver event N 120-n) occurring during driving that are associated with one or more driving values that are above one or more maximal driving thresholds or below one or more minimal driving thresholds. This can be depicted as values that are outside a Segment Trajectory Box (STB) 110 which is defined as the admissible values of the driver event parameters for a given road segment traveled by the driver driving the vehicle during the drive under certain driving and environment conditions. The STB 110 can be, for example, an area in six-dimensional space – having three measurements, two axes and two extremums. A given driver event (e.g., driver event A 120-a, driver event B 120-b, driver event C 120-c, …, driver event N 120-n) can be depicted as a point in this six-dimensional space. A given exemplary driver event (e.g., driver event A 120-a, driver event B 120-b, driver event C 120-c, …, driver event N 120-n) with its measured values being within the STB 110 is an admissible driver event. Another exemplary driver event (e.g., driver event A 120-a, driver event B 120-b, driver event C 120-c, …, driver event N 120-n) that has measured values being outside of the STB 110 is an unusual driver event. In some cases, the maximal driving thresholds and the minimal driving thresholds are dynamically calculated for at least part of the timeframe in which the driver is driving the vehicle, in accordance with one or more variables associated with driving the vehicle during the part of the timeframe. In some cases, the variables associated with driving the vehicle during part of the timeframe include one or more of: time of day, vehicle load, road geometry, and movement trajectory of the vehicle over the part of the timeframe, etc. In these cases, the dynamic calculations can be based on a second machine learning model, capable of receiving one or more variables describing a road segment that is traveled during the part of the timeframe by the driver driving the vehicle and determining the corresponding maximal driving thresholds and the corresponding minimal driving thresholds defining the STB 110. The driver events (e.g., driver event A 120-a, driver event B 120-b, driver event C 120-c, …, driver event N 120-n) can include one or more of: longitude velocity, latitude velocity, longitude acceleration, latitude acceleration, longitude jerk, latitude jerk, etc. 1. After obtaining the machine learning model and the given EDV of the given driver, system 200 can be further configured to determine one or more given values of the FBMs utilizing the machine learning model and the given EDV (block 320). It is to be noted that, in some cases, the given EDV is calculated based on the density of unusual driver events occurring during the given timeframe. The given EDV can be normalized over a distance driven during the given timeframe. After determining the values of the FBMs, system 200 is further configured to calculate the score for the given driver by comparing the given values of the FBMs to other values of FBMs of other drivers from the training dataset (block 330). It is to be noted that in some cases, the comparison of the score calculated for the given driver can be, for example, a comparison to an average score of a group of other drivers of the fleet. In some cases, system 200 is further configured to determine an explanation of the score calculated for the given driver, wherein the explanation is based on weights of the machine learning model. In a non-limiting example, a given drive is denoted as D. Drive D has a two-dimensional EDV (xd1=0.3, xd2=0.4). System 200 has defined a reference group of drives from the training dataset with a weighted average EDV (weighted by distance of each drive) of (xr1=0.2, xr2=0.1). The machine learning model in this example is a linear regression model (in the format of: ax1+bx2+c) with two corresponding coefficients a = 0.5, b=0.8, while the constant c is zero. Thus, the FBM for drive D is 0.3*0.5 + 0.4*0.8 = 0.47 and the FBM for the reference drives is 0.2*0.5+0.1*0.8 = 0.18. The difference between these two is 0.29. System 200 can explain this difference, by an impact analysis: impact1 = a(xd1-xr1) = 0.05 and impact2 = b(xd2-xr2) = 0.24 separately. System 200 can now determine that the impact of the second component is much higher than that of the first, so system 200 can provide an explanation of which part of the score (no matter how we calculate it) is due to which component. It is to be noted that in some cases, the machine learning model is a linear Non-Negative Least Squares (NNLS) regression model. It is to be noted that, with reference to Fig. 3, some of the blocks can be integrated into a consolidated block or can be broken down to a few blocks and/or other blocks may be added. Furthermore, in some cases, the blocks can be performed in a different order than described herein. It is to be further noted that some of the blocks are optional. It should also be noted that whilst the flow diagram is described also with reference to the system elements that realizes them, this is by no means binding, and the blocks can be performed by elements other than those described herein. Fig. 4 is a flowchart illustrating one example of a sequence of operations carried out for driving thresholds determination for a road segment, in accordance with the presently disclosed subject matter. According to certain examples of the presently disclosed subject matter, system 200 can be configured to perform a driving thresholds determination process 400, e.g., utilizing the driving thresholds determination module 250. A Segment Trajectory Box (STB) 110 is defined as the admissible values of the driver event parameters for a given road segment traveled by the driver driving the vehicle during the drive under certain driving and environment conditions. The STB 110 can be, for example, an area in six-dimensional space – having three measurements, two axes and two extremums. A given driver event (e.g., driver event A 120-a, driver event B 120-b, driver event C 120-c, …, driver event N 120-n) can be depicted as a point in this six-dimensional space. A given exemplary driver event that in accordance with its measured values being within the STB 110 is an admissible driver event. Another exemplary driver event that has measured values that are outside of the STB 110 is an unusual driver event. The STB 110 can be dynamically determined by the system, for example: an STB 110 is determined for every segment of the road driven during the drive. A segment of the road has a certain geometry, and a driver needs to decide on a specific trajectory for passing the road segment given context and/or environment variables. The context variable can include, for example: time of day, vehicle load, etc. The segment's geometry can be for example: a left turn of a specific curvature. The trajectory taken by the driver for that segment under those conditions can be described as movement along the geometry as a function of time. A trajectory can be described by its extremum values – maximum and minimum of velocity, acceleration and jerk over two axes of longitude and latitude. As mentioned above, the dynamically determined STB 110 can be used to determine if a driver event (e.g., driver event A 120-a, driver event B 120-b, driver event C 120-c, …, driver event N 120-n) is an admissible driver event with a usual trajectory by looking at the driver events (e.g., driver event A 120-a, driver event B 120-b, driver event C 120-c, …, driver event N 120-n) as an input defined by twelve values is within the STB 110 in six-dimensional space. Unusual driver events occur when a specific trajectory is outside of the admissible six-dimensional STB 110. Optionally and/or additionally, the system can measure the magnitude of the driver event – how far is that given driver event from the STB 110 and/or the duration in which the given driver event is out of the STB 110. In some cases, the described system can determine the STB 110 by utilizing a machine learning model capable of predicting a corresponding STB 110 for a given road segment. The machine learning model can be trained using historical data. A non-limiting example of such a machine learning model is a regression model trained to predict STB 110 from a segment's contextual variable and geometry representation. An average or usual STB 110 can be determined based on one or more historical drives by drivers passing through this segment and/or similar segments under similar conditions. In some cases, the geometry representation can be normalized by the system so that similar segment geometries (for example: straight segments or sharp left-turn segments) will have similar representations. In these cases, it can be assumed that similar geometry representations having similar contextual variables will produce similar STBs 110. For this purpose, system 200 can be configured to obtain: (A) a machine learning model capable of receiving: (i) one or more road segment variables describing the road segment that is traveled during a timeframe, and (ii) one or more driving conditions variables describing driving conditions associated with an area of the road segment during the timeframe, and determining the driving thresholds for the given road segment, (B) one or more given road segment variables of a given road segment that is traveled during a given timeframe, and (C) one or more given driving conditions variables associated with a given area of the given road segment during the given timeframe (block 410). It is to be noted that the term obtain includes: obtaining the road segment variables and/or the driving conditions variables from the environment of the given road (for example: by utilizing a database describing of the given road and/or utilizing sensors sensing the given road), obtaining the road segment variables and/or the driving conditions variables from an external system, external to system 200 and/or generating the road segment variables and/or the driving conditions variables by system 200 itself. After obtaining the machine learning model, the given road segment variables and the given driving condition variables, system 200 can be further configured to determine the driving thresholds for the given road segment by utilizing the machine learning model and the given road segment variables (block 420). It is to be noted that, with reference to Fig. 4, some of the blocks can be integrated into a consolidated block or can be broken down to a few blocks and/or other blocks may be added. Furthermore, in some cases, the blocks can be performed in a different order than described herein. It is to be further noted that some of the blocks are optional. It should also be noted that whilst the flow diagram is described also with reference to the system elements that realizes them, this is by no means binding, and the blocks can be performed by elements other than those described herein. It is to be understood that the presently disclosed subject matter is not limited in its application to the details set forth in the description contained herein or illustrated in the drawings. The presently disclosed subject matter is capable of other embodiments and of being practiced and carried out in various ways. Hence, it is to be understood that the phraseology and terminology employed herein are for the purpose of description and should not be regarded as limiting. As such, those skilled in the art will appreciate that the conception upon which this disclosure is based may readily be utilized as a basis for designing other structures, methods, and systems for carrying out the several purposes of the presently disclosed subject matter. It will also be understood that the system according to the presently disclosed subject matter can be implemented, at least partly, as a suitably programmed computer. Likewise, the presently disclosed subject matter contemplates a computer program being readable by a computer for executing the disclosed method. The presently disclosed subject matter further contemplates a machine-readable memory tangibly embodying a program of instructions executable by the machine for executing the disclosed method.

Claims (24)

- 27 - CLAIMS:

1. A system for determining a score for a driver of a vehicle of a fleet of vehicles, the score is determined with respect to one or more Fleet Business Metrics (FBMs) of the fleet of vehicles, the system comprising a processing circuitry configured to: obtain: (A) a machine learning model capable of receiving an Event Density Vector (EDV) indicative of a density of unusual driver events occurring during a timeframe in which the driver is driving the vehicle and determining one or more values of the FBMs, wherein the machine learning model is trained utilizing a labeled training-data set comprising of a plurality of training records, each training record comprising: (i) a training EDV, indicative of the density of unusual driver events occurring during a given training drive on a training vehicle, and (ii) one or more training values of the FBMs corresponding to the given training drive, wherein at least one of the training values of the FBMs are calculated based on sensor readings captured from the training vehicle, and (B) a given EDV of a given driver driving the vehicle during a given timeframe; determine one or more given values of the FBMs utilizing the machine learning model and the given EDV; and calculate the score for the given driver by comparing the given values of the FBMs to other values of FBMs of other drivers from the training-data set.

2. The system of claim 1, wherein the unusual driver events are driver events occurring during driving that are associated with one or more driving values that are above one or more maximal driving thresholds or below one or more minimal driving thresholds. - 28 -

3. The system of claim 2, wherein the maximal driving thresholds and the minimal driving thresholds are dynamically calculated for at least part of the timeframe in which the driver is driving the vehicle, in accordance with one or more variables associated with driving the vehicle during the part of the timeframe.

4. The system of claim 3, wherein the variables associated with driving the vehicle during the part of the timeframe include one or more of: time of day, vehicle load, road geometry, and movement trajectory of the vehicle over the part of the timeframe.

5. The system of claim 3, wherein the dynamic calculations are based on a second machine learning model, capable of receiving one or more variables describing a road segment that is traveled during the part of the timeframe by the driver driving the vehicle and determining the corresponding maximal driving thresholds and the corresponding minimal driving thresholds.

6. The system of claim 2, wherein the driver events include one or more of: longitude velocity, latitude velocity, longitude acceleration, latitude acceleration, longitude jerk, and latitude jerk.

7. The system of claim 1, wherein the given EDV is calculated based on the density of unusual driver events occurring during the given timeframe.

8. The system of claim 7, wherein the given EDV is normalized over a distance driven during the given timeframe.

9. The system of claim 1, wherein the processing circuitry is further configured to determine an explanation of the score calculated for the given driver, wherein the explanation is based on weights of the machine learning model.

10. The system of claim 1, wherein the machine learning model is a linear Non- Negative Least Squares (NNLS) regression model. - 29 -

11. A system for determining one or more driving thresholds for a road segment, the system comprising a processing circuitry configured to: obtain: (A) a machine learning model capable of receiving: (i) one or more road segment variables describing the road segment that is traveled during a timeframe, and (ii) one or more driving conditions variables describing driving conditions associated with an area of the road segment during the timeframe, and determining the driving thresholds for the given road segment, (B) one or more given road segment variables of a given road segment that is traveled during a given timeframe, and (C) one or more given driving conditions variables associated with a given area of the given road segment during the given timeframe; and determine the driving thresholds for the given road segment by utilizing the machine learning model and the given road segment variables.

12. A method for determining a score for a driver of a vehicle of a fleet of vehicles, the score is determined with respect to one or more Fleet Business Metrics (FBMs) of the fleet of vehicles, the method comprising: obtain, using a processing circuitry: (A) a machine learning model capable of receiving an Event Density Vector (EDV) indicative of a density of unusual driver events occurring during a timeframe in which the driver is driving the vehicle and determining one or more values of the FBMs, wherein the machine learning model is trained utilizing a labeled training-data set comprising of a plurality of training records, each training record comprising: (i) a training EDV, indicative of the density of unusual driver events occurring during a given training drive on a training vehicle, and (ii) one or more training values of the FBMs corresponding to the given training drive, wherein at least one of the training values of the FBMs are calculated based on sensor readings captured from the training vehicle, and (B) a given EDV of a given driver driving the vehicle during a given timeframe; - 30 - determine, using the processing circuitry, one or more given values of the FBMs utilizing the machine learning model and the given EDV; and calculate, using the processing circuitry, the score for the given driver by comparing the given values of the FBMs to other values of FBMs of other drivers from the training-data set.

13. The method of claim 12, wherein the unusual driver events are driver events occurring during driving that are associated with one or more driving values that are above one or more maximal driving thresholds or below one or more minimal driving thresholds.

14. The method of claim 13, wherein the maximal driving thresholds and the minimal driving thresholds are dynamically calculated for at least part of the timeframe in which the driver is driving the vehicle, in accordance with one or more variables associated with driving the vehicle during the part of the timeframe.

15. The method of claim 14, wherein the variables associated with driving the vehicle during the part of the timeframe include one or more of: time of day, vehicle load, road geometry, and movement trajectory of the vehicle over the part of the timeframe.

16. The method of claim 14, wherein the dynamic calculations are based on a second machine learning model, capable of receiving one or more variables describing a road segment that is traveled during the part of the timeframe by the driver driving the vehicle and determining the corresponding maximal driving thresholds and the corresponding minimal driving thresholds.

17. The method of claim 13, wherein the driver events include one or more of: longitude velocity, latitude velocity, longitude acceleration, latitude acceleration, longitude jerk, and latitude jerk.

18. The method of claim 12, wherein the given EDV is calculated based on the density of unusual driver events occurring during the given timeframe. - 31 -

19. The method of claim 18, wherein the given EDV is normalized over a distance driven during the given timeframe.

20. The method of claim 12, further configured to determine, using the processing circuitry, an explanation of the score calculated for the given driver, wherein the explanation is based on weights of the machine learning model.

21. The method of claim 12, wherein the machine learning model is a linear Non-Negative Least Squares (NNLS) regression model.

22. A method for determining one or more driving thresholds for a road segment, the method comprising: obtain, using a processing circuitry: (A) a machine learning model capable of receiving: (i) one or more road segment variables describing the road segment that is traveled during a timeframe, and (ii) one or more driving conditions variables describing driving conditions associated with an area of the road segment during the timeframe, and determining the driving thresholds for the given road segment, (B) one or more given road segment variables of a given road segment that is traveled during a given timeframe, and (C) one or more given driving conditions variables associated with a given area of the given road segment during the given timeframe; and determine, using the processing circuitry, the driving thresholds for the given road segment by utilizing the machine learning model and the given road segment variables. - 32 -

23. A non-transitory computer readable storage medium having computer readable program code embodied therewith, the computer readable program code, executable by processing circuitry of a computer to perform a method comprising: obtain, using a processing circuitry: (A) a machine learning model capable of receiving an Event Density Vector (EDV) indicative of a density of unusual driver events occurring during a timeframe in which the driver is driving the vehicle and determining one or more values of the FBMs, wherein the machine learning model is trained utilizing a labeled training-data set comprising of a plurality of training records, each training record comprising: (i) a training EDV, indicative of the density of unusual driver events occurring during a given training drive on a training vehicle, and (ii) one or more training values of the FBMs corresponding to the given training drive, wherein at least one of the training values of the FBMs are calculated based on sensor readings captured from the training vehicle, and (B) a given EDV of a given driver driving the vehicle during a given timeframe; determine, using the processing circuitry, one or more given values of the FBMs utilizing the machine learning model and the given EDV; and calculate, using the processing circuitry, the score for the given driver by comparing the given values of the FBMs to other values of FBMs of other drivers from the training-data set.

24. A non-transitory computer readable storage medium having computer readable program code embodied therewith, the computer readable program code, executable by processing circuitry of a computer to perform a method comprising: obtain, using a processing circuitry: (A) a machine learning model capable of receiving: (i) one or more road segment variables describing the road segment that is traveled during a timeframe, and (ii) one or more driving conditions variables describing driving conditions associated with an area of the road segment during the timeframe, and determining the driving thresholds for the given road segment, - 33 - (B) one or more given road segment variables of a given road segment that is traveled during a given timeframe, and (C) one or more given driving conditions variables associated with a given area of the given road segment during the given timeframe; and determine, using the processing circuitry, the driving thresholds for the given road segment by utilizing the machine learning model and the given road segment variables.