TWI835176B - Bloodstream infection predicting system and method thereof - Google Patents

Abstract

A bloodstream infection predicting system is proposed. The bloodstream infection predicting system includes a memory unit and a processor. The memory unit stores a plurality of historical medical data, a real-time data to be tested and a machine learning algorithm. The processor is configured to implement the above steps. The first data reading step is performed to read the historical medical data. The model training step is performed to train the historical medical data to generate a bloodstream infection predicting model. The second data reading step is performed to read the real-time data to be tested of a patient. The risk predicting step is performed to input the real-time data to be tested to the bloodstream infection predicting model to generate a bloodstream infection risk probability. The real-time data to be tested includes an intensive care unit (ICU) detecting data which is detected during a feature window period and a hematology feature data. Thus, the bloodstream infection predicting system of the present disclosure can predict the bloodstream infection risk probability after a specific time, and the medical treatment can be given immediately.

Description

Bacteremia infection risk prediction system and method thereof

The present invention relates to an infection risk prediction system and a method thereof, in particular to a bacteremia infection risk prediction system and a method thereof.

Bacteremia is a common disease in the intensive care unit. However, bacteremia cannot be diagnosed immediately. Medical staff need to conduct bacterial culture on the blood of patients with bacteremia. Only after the bacterial culture is completed can the patient be confirmed to have bacteremia. , and provide corresponding medical treatment. Since bacteremia is the main complication and cause of death in critically ill patients, and the mortality rate of patients with bacteremia is extremely high, waiting for bacterial culture time before providing medical treatment to the patient is likely to miss the best treatment time.

It can be seen that there is still a lack of a bacteremia infection risk prediction system and method on the market that can instantly monitor the health of ICU patients and predict the risk of bacteremia infection. This is indeed something that the public is eagerly waiting for, and it is also the reason why relevant industry players must work hard. The goal and direction of R&D breakthroughs.

Therefore, the purpose of the present invention is to provide a bacteremia infection risk prediction system and a method thereof, which predict the patient's risk probability of bacteremia infection after a specific period of time by collecting the patient's real-time test data.

According to one embodiment of the structural aspect of the present invention, a bacteremia infection risk prediction system is provided, which is used to predict a bacteremia infection risk probability based on a patient's real-time test data. The bacteremia infection risk prediction system includes a memory unit and a processor. The memory unit stores multiple historical medical data, real-time data to be tested and a machine learning algorithm. The processor signal is connected to the memory unit and is configured to implement a first data reading step, a model training step, a second data reading step and a risk prediction step. The first data reading step is to read the historical medical data in the memory unit. The model training step is to train these historical medical data based on a machine learning algorithm to generate a bacteremia prediction model. The second data reading step is to read the patient's real-time data to be tested from the memory unit. The risk prediction step is to input the real-time test data into the bacteremia prediction model to generate the risk probability of bacteremia infection. The real-time data to be measured includes an intensive care unit monitoring information and a blood test characteristic information of the patient within a characteristic window interval.

Thereby, the bacteremia infection risk prediction system of the present invention predicts the probability of a patient being infected with bacteremia after a specific period of time and provides early medical treatment.

Other examples of the aforementioned implementation are as follows: the aforementioned memory unit stores a preset quantity and a preset quantity lower limit, and each historical medical data includes plural feature information. The processor is further configured to perform a data pre-processing step. The data pre-processing step includes the driver processor calculating an average value of each characteristic information and the driver processor determining whether the quantity of these characteristic information of each historical medical data is less than or equal to a preset quantity lower limit. When the quantity of the characteristic information of one of the historical medical data is less than or equal to the preset quantity lower limit, the processor removes this one of the historical medical data. When the quantity of the characteristic information of the historical medical data is greater than the preset quantity lower limit and less than the preset quantity, the processor adds the missing part of the characteristics of the historical medical data according to an interpolation procedure. The average values corresponding to the information are filled in the historical medical data, so that the number of characteristic information in the historical medical data is equal to the preset number.

Other examples of the aforementioned embodiments are as follows: the aforementioned intensive care unit monitoring information includes a body temperature, a respiratory rate, a pulse rate, a pulse pressure, a systolic blood pressure, a diastolic blood pressure, a coma index and an intubation time information.

Other examples of the aforementioned embodiments are as follows: the aforementioned blood test characteristic information includes a lactate value, an arterial blood gas pH value and a bicarbonate value.

Other examples of the aforementioned implementation are as follows: the aforementioned machine learning algorithm is one of a logistic regression, a support vector machine, a multi-layer perceptron, a random forest algorithm and an extreme gradient boosting algorithm.

According to one embodiment of the method aspect of the present invention, a bacteremia infection risk prediction method is provided, which is used to predict a bacteremia infection risk probability based on a patient's real-time test data. The bacteremia infection risk prediction method includes a first data reading step, a model training step, a second data reading step and a risk prediction step. The first data reading step is to drive a processor to read a plurality of historical medical data in a memory unit. The model training step drives the processor to train the historical medical data according to a machine learning algorithm to generate a bacteremia prediction model. The second data reading step is to drive the processor to read the patient's real-time data to be tested from the memory unit. The risk prediction step drives the processor to input real-time data to be tested into the bacteremia prediction model to generate a bacteremia infection risk probability. The real-time data to be measured includes an intensive care unit monitoring information and a blood test characteristic information of the patient within a characteristic window interval.

Thereby, the bacteremia infection risk prediction method of the present invention predicts the probability of a patient being infected with bacteremia after a specific period of time and provides early medical treatment.

Other examples of the aforementioned implementation are as follows: the aforementioned memory unit stores a preset quantity and a preset quantity lower limit, and each historical medical data includes plural feature information. The bacteremia infection risk prediction method further includes a data pre-processing step. The data pre-processing step includes the driver processor calculating an average value of each characteristic information and the driver processor determining whether the quantity of these characteristic information of each historical medical data is less than or equal to a preset quantity lower limit. When the quantity of the characteristic information of one of the historical medical data is less than or equal to the preset quantity lower limit, the processor removes this one of the historical medical data. When the quantity of the characteristic information of the historical medical data is greater than the preset quantity lower limit and less than the preset quantity, the processor adds the missing part of the characteristics of the historical medical data according to an interpolation procedure. The average values corresponding to the information are filled in the historical medical data, so that the number of characteristic information in the historical medical data is equal to the preset number.

Other examples of the aforementioned embodiments are as follows: the aforementioned intensive care unit monitoring information includes a body temperature, a respiratory rate, a pulse rate, a pulse pressure, a systolic blood pressure, a diastolic blood pressure, a coma index and an intubation time information.

Other examples of the aforementioned embodiments are as follows: the aforementioned blood test characteristic information includes a lactate value, an arterial blood gas pH value and a bicarbonate value.

Other examples of the aforementioned implementation are as follows: the aforementioned machine learning algorithm is one of a logistic regression, a support vector machine, a multi-layer perceptron, a random forest algorithm and an extreme gradient boosting algorithm.

Several embodiments of the present invention will be described below with reference to the drawings. For the sake of clarity, many practical details will be explained together in the following narrative. However, it will be understood that these practical details should not limit the invention. That is to say, in some embodiments of the present invention, these practical details are not necessary. In addition, in order to simplify the drawings, some commonly used structures and components will be illustrated in a simple schematic manner in the drawings; and repeated components may be represented by the same numbers.

In addition, when a certain component (or unit or module, etc.) is "connected" to another component in this article, it may mean that the component is directly connected to the other component, or it may mean that one component is indirectly connected to the other component. , meaning that there are other elements between the said element and another element. When it is stated that an element is "directly connected" to another element, it means that no other elements are interposed between the element and the other element. Terms such as first, second, third, etc. are only used to describe different components without limiting the components themselves. Therefore, the first component can also be renamed the second component. Moreover, the combination of components/units/circuit in this article is not a combination that is generally known, conventional or customary in this field. Whether the component/unit/circuit itself is common knowledge cannot be used to determine whether its combination relationship is easily understood by those in the technical field. Usually it is easily accomplished by the knowledgeable.

Please refer to Figures 1 and 2. Figure 1 is a block diagram illustrating the bacteremia infection risk prediction system 100 according to the first embodiment of the present invention; Figure 2 is a schematic block diagram illustrating the bacteremia infection risk prediction system 100 according to the embodiment of Figure 1. A schematic diagram of the real-time test data 221 of the blood infection risk prediction system 100. The bacteremia infection risk prediction system 100 is used to predict a bacteremia infection risk probability 320 based on a patient's real-time test data 221 . The bacteremia infection risk prediction system 100 includes a memory unit 200 and a processor 300 . The memory unit 200 stores a plurality of historical medical data 211, real-time data to be tested 221 and a machine learning algorithm 230. The real-time data to be tested 221 includes an intensive care unit monitoring information 2211 and a blood test characteristic information 2212 of the patient in a characteristic window interval T12 (see Figure 4).

Specifically, the memory unit 200 stores a historical database 210, a real-time database 220 and a machine learning algorithm 230. The memory unit 200 may be a memory or other data storage device. The historical database 210 contains historical medical data 211 . Each historical medical data 211 includes historical characteristic information of patients who have been admitted to the intensive care unit. The real-time database 220 includes real-time collected intensive care unit monitoring information 2211 and blood test characteristic information 2212 of the patient to be tested within the characteristic window interval T12. In this embodiment, the real-time database 220 stores the real-time data 221 of the patient to be tested in the intensive care unit within 72 hours (ie, the characteristic window interval T12).

In detail, the real-time data to be tested 221 includes intensive care unit monitoring information 2211 monitored when the patient is in the intensive care unit, blood test characteristic information 2212 and other characteristic information. The intensive care unit monitoring information 2211 may include a temperature (Temperature), a respiration rate (Respiration Rate), a pulse rate (Pulse Rate), a pulse pressure (Pulse Pressure), a systolic blood pressure, a diastolic blood pressure, and a coma index (GCS). ) and an intubation time information. Intubation time information may include pulmonary artery catheter (SwanGanze) intubation time, intubation (ENDO) time, urinary catheter (Foley) intubation time, central venous catheter (Central Venous Catheter; CVC) intubation time, central venous pressure (Central Venous Pressure) intubation time, Double Lumen (Double Lumen) intubation time, Hickman Catheter (Hickman Catheter) intubation time, Peripherally Inserted Central Catheters (PICC) intubation time and implantable intravenous catheter (Port A) intubation time, but the present invention is not limited thereto.

The blood test characteristic information 2212 may include a lactate value (Lactate), an arterial blood gas pH value (Arterial Blood Gas_pH), a bicarbonate value (HCO ₃ -A), a white blood cell count (WBC-min), a blood Medium urea nitrogen (BUN), alkaline phospholipase (Alkaline Phosphatase; ALKP), heme (Hb), sodium (K), creatinine (Creatinine) and prothrombin time- C). In this embodiment, other characteristic information may include intensive care unit admission time and severity evaluation index (Acute Physiology and Chronic Health Evaluation; APACHE II score), but the invention is not limited thereto.

The processor 300 may be a microprocessor, a central processing unit (Central Processing Unit; CPU), or other electronic computing processor, but the invention is not limited thereto. The processor 300 is connected to the memory unit 200 via signals, and is configured to implement a first data reading step S02, a model training step S04, a second data reading step S06 and a risk prediction step S08. The first data reading step S02 is to read the historical medical data 211 of the memory unit 200 . The model training step S04 is based on the machine learning algorithm 230 to train the historical medical data 211 to generate a bacteremia prediction model 310. The second data reading step S06 is to read the patient's real-time data to be tested 221 from the memory unit 200 . The risk prediction step S08 is to input the real-time test data 221 into the bacteremia prediction model 310 to generate a bacteremia infection risk probability 320. Thereby, the bacteremia infection risk prediction system 100 of the present invention collects the characteristic parameters monitored in the intensive care unit that are highly related to bacteremia infection, and then predicts the patient's bacteremia after a specific time period T23 (see Figure 4). The infection risk probability is 320, and the patient should be given early medical treatment before the bacterial culture of bacteremia is completed. The following will describe the operations of the first data reading step S02, the model training step S04, the second data reading step S06 and the risk prediction step S08 through a more detailed embodiment.

Please refer to Figures 1 to 4. Figure 3 is a flow chart illustrating the bacteremia infection risk prediction method S10 according to the second embodiment of the present invention; Figure 4 is a flow chart illustrating the bacteremia infection risk prediction method according to the embodiment of Figure 3. Schematic diagram of the characteristic window interval T12 of the blood infection risk prediction method S10. The bacteremia infection risk prediction method S10 is used to predict the bacteremia infection risk probability 320 based on the patient's real-time test data 221 . The bacteremia infection risk prediction method S10 includes a first data reading step S02, a model training step S04, a second data reading step S06 and a risk prediction step S08.

In Figure 4, time t0 on the timeline t represents the time point when the patient enters the intensive care unit, the time between time t1 and time t2 is the characteristic window interval T12, and the time between time t2 and time t3 is the specific time. Section T23.

The first data reading step S02 is to drive the processor 300 to read the plurality of historical medical data 211 in the memory unit 200 .

The model training step S04 is to drive the processor 300 to train the historical medical data 211 according to the machine learning algorithm 230 to generate a bacteremia prediction model 310. Specifically, the machine learning algorithm 230 may be one of a logistic regression, a support vector machine, a multi-layer perceptron, a random forest algorithm, and an extreme gradient boosting algorithm, but the present invention does not take this as an example. limit.

The second data reading step S06 is to drive the processor 300 to read the patient's real-time data to be tested 221 from the memory unit 200 . Specifically, the second data reading step S06 is to drive the processor 300 to read the real-time test data 221 of the patient within the characteristic window interval T12 stored in the real-time database 220 of the memory unit 200. In this embodiment, the The second data reading step S06 reads the patient's real-time test data 221 from time t1 to time t2 at time t2. The characteristic window interval T12 is 72 hours, that is, the second data reading step S06 reads the patient's real-time data to be tested 221 from time t1 to time t2. All the real-time test data 221 of the tested patient from 72 hours ago (i.e., time t1) to the current time t2, but the present invention is not limited to this.

The risk prediction step S08 drives the processor 300 to input the real-time test data 221 into the bacteremia prediction model 310 to generate a bacteremia infection risk probability 320. The real-time data to be tested 221 includes the patient's intensive care unit monitoring information 2211 and blood test characteristic information 2212 in the characteristic window interval T12. Specifically, the bacteremia prediction model 310 is used to predict the bacteremia infection risk probability 320 of a patient diagnosed with bacteremia at time t3. In this embodiment, the specific time period T23 between time t3 and time t2 is 24 hours, but the invention is not limited thereto.

Thereby, the bacteremia infection risk prediction method S10 of the present invention continues to read the patient's real-time test data 221 in the characteristic window interval T12 through the second data reading step S06, and predicts the patient's specific time period through the risk prediction step S08. The risk probability of bacteremia infection after T23 is 320, and an immediate warning reminder is generated for medical staff in the intensive care unit to provide timely and accurate medical treatment.

The bacteremia infection risk prediction method S10 further includes a data preprocessing step S01. Each historical medical data 211 includes plural feature information. The memory unit 200 stores a preset quantity and a preset quantity lower limit. The data preprocessing step S01 includes driving the processor 300 to calculate an average value of each feature information and driving the processor 300 to determine whether the number of the feature information of each historical medical data 211 is less than or equal to a preset lower limit.

In detail, the characteristic information of each historical medical data 211 corresponds to the intensive care unit monitoring information 2211 and blood test characteristic information 2212 of the real-time data to be tested 221 . The data preprocessing step S01 calculates the average value of each characteristic information in the historical medical data 211 .

When the quantity of the characteristic information of one of the historical medical data 211 is less than or equal to the preset quantity lower limit, the processor 300 removes this one of the historical medical data 211 . In other words, the data preprocessing step S01 is used to confirm whether the characteristic information in each piece of historical medical data 211 is missing. At the same time, when the amount of missing characteristic information is too much, the historical medical data 211 is removed. This historical medical data 211 is used as a training sample for the bacteremia prediction model 310. In this embodiment, when the characteristic information of each historical medical data 211 is not missing, the number of these characteristic information is equal to the preset number. The default number of characteristic information of each historical medical data 211 is 20. The default The lower limit of the quantity can be 60% of the default quantity of these feature information (ie, 12 feature information). When the missing feature information in a piece of historical medical data 211 is greater than 40% (ie, 8 feature information), the historical medical data 211 will be removed, but the present invention is not limited to this.

When the quantity of the characteristic information of the historical medical data 211 is greater than the preset quantity lower limit and less than the preset quantity, the processor 300 adds the missing part of the historical medical data 211 according to an interpolation procedure. The average values corresponding to the characteristic information are filled in the historical medical data 211, so that the number of the characteristic information in the historical medical data 211 is equal to the preset quantity. Specifically, when a small amount of feature information is missing in a historical medical data 211, the data preprocessing step S01 brings the average value corresponding to the missing feature information into the historical medical data 211 for model training. In this embodiment, when one of the characteristic information of a piece of historical medical data 211 is missing, the data preprocessing step S01 calculates the average of the missing characteristic information of all historical medical data 211, and calculates the average value of the missing characteristic information. The average value is filled in this historical medical data 211.

Thereby, the bacteremia infection risk prediction method S10 of the present invention filters the incomplete historical medical data 211 through the data preprocessing step S01, reduces the prediction error value of the bacteremia prediction model 310, and increases the bacteremia infection risk probability 320 The accuracy rate can thereby reduce the bacteremia infection rate in the ICU, improve the quality of care, and shorten the length of stay of patients in the ICU.

It can be seen from the above embodiments that the bacteremia infection risk prediction system and method of the present invention have the following advantages. First, it collects characteristic parameters monitored in the intensive care unit that are highly related to bacteremia infection, and then predicts the risk of patients in a specific time period. The risk probability of bacteremia infection in the future is determined, and medical treatment is given to the patient as early as possible before the bacterial culture of bacteremia is completed; secondly, the patient's real-time test data in the characteristic window interval is continuously read through the second data reading step, and the The risk prediction step predicts the patient's risk probability of bacteremia infection after a specific period of time, and generates an immediate warning reminder to provide timely and accurate medical treatment to the medical staff in the intensive care unit; thirdly, the filtering through the data pre-processing step is incomplete. historical medical data, reduce the prediction error of the bacteremia prediction model, and improve the accuracy of the risk probability of bacteremia infection, thus reducing the bacteremia infection rate in the ICU, improving the quality of care, and shortening the patient's hospitalization in the ICU Number of days.

Although the present invention has been disclosed in the above embodiments, it is not intended to limit the present invention. Anyone skilled in the art can make various modifications and modifications without departing from the spirit and scope of the present invention. Therefore, the protection of the present invention is The scope shall be determined by the appended patent application scope.

100:Bacteremia infection risk prediction system 200:Memory unit 210:Historical database 211: Historical medical data 220:Real-time database 221: Real-time data to be tested 2211: Intensive care unit monitoring information 2212: Blood test characteristics information 230:Machine Learning Algorithm 300:processor 310:Bacteremia prediction model 320:Bacteremia infection risk probability S01: Data pre-processing steps S02: First data reading step S04: Model training steps S06: Second data reading step S08: Risk prediction step S10: Methods for predicting the risk of bacteremia infection t: time axis t0,t1,t2,t3: time T12: Feature window interval T23: specific time period

Figure 1 is a block diagram illustrating a bacteremia infection risk prediction system according to the first embodiment of the present invention; Figure 2 is a schematic diagram illustrating real-time data to be tested in the bacteremia infection risk prediction system according to the implementation of Figure 1; Figure 3 is a flow chart illustrating a bacteremia infection risk prediction method according to the second embodiment of the present invention; and Figure 4 is a schematic diagram illustrating the characteristic window interval of the bacteremia infection risk prediction method according to the embodiment of Figure 3.

100:Bacteremia infection risk prediction system

200:Memory unit

210:Historical database

211: Historical medical data

220:Real-time database

221: Real-time data to be tested

230:Machine Learning Algorithm

300:processor

310:Bacteremia prediction model

320:Bacteremia infection risk probability

S02: First data reading step

S04: Model training steps

S06: Second data reading steps

S08: Risk prediction step

Claims (6)

A bacteremia infection risk prediction system is used to predict a bacteremia infection risk probability based on real-time test data of a patient. The bacteremia infection risk prediction system includes: a memory unit that stores a plurality of historical medical data, the Real-time data to be tested and a machine learning algorithm; and a processor, the signal is connected to the memory unit, and is configured to implement the following steps: a first data reading step, which is to read the historical medical treatments of the memory unit data; a model training step, which is based on the machine learning algorithm to train the historical medical data to generate a bacteremia prediction model; a second data reading step, which is to read the patient's immediate treatment from the memory unit The test data; and a risk prediction step is to input the real-time test data into the bacteremia prediction model to generate the risk probability of bacteremia infection; wherein the real-time test data includes the patient's risk probability within a characteristic window interval. An intensive care unit monitoring information and a blood test characteristic information; wherein, the second data reading step continuously reads the real-time test data of the patient in the characteristic window interval, and predicts the patient in a specific period through the risk prediction step. The bacteremia infection risk probability after the time period; wherein the blood test characteristic information includes a lactate value, an arterial blood gas pH value and a bicarbonate value; wherein the intensive care unit monitoring information includes an intubation time Information should be inserted The catheter time information includes an intubation time, a urinary catheter intubation time and a central venous catheter intubation time; wherein the memory unit stores a preset quantity and a preset quantity lower limit, and each historical medical data includes plural characteristic information, And the processor is further configured to implement: a data pre-processing step, including: driving the processor to calculate an average value of each of the characteristic information; and driving the processor to determine the quantity of the characteristic information of each of the historical medical data. Whether it is less than or equal to the preset lower limit of quantity; wherein, when the quantity of the characteristic information of one of the historical medical data is less than or equal to the preset lower limit of quantity, the processor removes that one of the historical medical data; And wherein, when the quantity of the characteristic information of the historical medical data is greater than the preset quantity lower limit and less than the preset quantity, the processor uses an interpolation procedure to add the quantity of the historical medical data to the The missing part of the characteristic information corresponding to the average value is filled in the historical medical data, so that the quantity of the characteristic information in the historical medical data is equal to the preset quantity. The bacteremia infection risk prediction system as described in claim 1, wherein the intensive care unit monitoring information further includes a body temperature, a respiratory rate, a pulse rate, a pulse pressure, a systolic blood pressure, a diastolic blood pressure and a coma index. Bacteremia infection risk prediction system as described in request 1, The machine learning algorithm is one of a logistic regression, a support vector machine, a multi-layer perceptron, a random forest algorithm and an extreme gradient boosting algorithm. A bacteremia infection risk prediction method is used to predict a bacteremia infection risk probability based on real-time test data of a patient. The bacteremia infection risk prediction method includes: a first data reading step, which drives a The processor reads a plurality of historical medical data from a memory unit; a model training step drives the processor to train the historical medical data according to a machine learning algorithm to generate a bacteremia prediction model; and a second data read A fetching step is to drive the processor to read the real-time data to be tested of the patient from the memory unit; and a risk prediction step is to drive the processor to input the real-time data to be tested into the bacteremia prediction model to generate the The risk probability of bacteremia infection; wherein, the real-time data to be measured includes an intensive care unit monitoring information and a blood test characteristic information of the patient in a characteristic window interval; wherein, the second data reading step continuously reads the patient's The real-time test data in the characteristic window interval is used to predict the patient's risk probability of bacteremia infection after a specific period of time through the risk prediction step; wherein the blood test characteristic information includes a lactate value, an arterial blood Gas pH value and a bicarbonate value; wherein, the intensive care unit monitoring information includes an intubation time information, and the intubation time information includes an intubation time, a urinary catheter intubation time and a central static Vascular catheter intubation time; wherein the memory unit stores a preset quantity and a preset quantity lower limit, each of the historical medical data includes plural feature information, and the bacteremia infection risk prediction method further includes: a data pre-processing step, It includes: driving the processor to calculate an average value of each feature information; and driving the processor to determine whether the quantity of the feature information of each historical medical data is less than or equal to the preset lower limit of quantity; wherein, when the historical medical data When the quantity of the characteristic information of one of the data is less than or equal to the preset quantity lower limit, the processor removes the characteristic of the historical medical data; and wherein, when the characteristic of the historical medical data When the amount of information is greater than the preset lower limit and less than the preset amount, the processor fills in the missing portion of the historical medical data with the average values corresponding to the feature information based on an interpolation procedure. Enter the historical medical data so that the number of characteristic information of the historical medical data is equal to the preset number. The bacteremia infection risk prediction method described in claim 4, wherein the intensive care unit monitoring information further includes a body temperature, a respiratory rate, a pulse rate, a pulse pressure, a systolic blood pressure, a diastolic blood pressure and a coma index. The bacteremia infection risk prediction method as described in claim 4, wherein the machine learning algorithm is a logistic regression, a support vector machine, One of a multilayer perceptron, a random forest algorithm, and an extreme gradient boosting algorithm.

Images