JP7386203B2 - Information processing device and information processing method - Google Patents

Description

Article 30, Paragraph 2 of the Patent Act applies (1) Published by Yoshinori Shiro, Fumie Nakaya, Daisuke Nishiyama, and Yoshito Kurihara on October 28, 2020 (2) Sales supervisor on November 24, 2020 Headquarters released

The present invention relates to an information processing device and an information processing method.

In recent years, technological development to promote digital transformation (DX) has been active, and as part of this, monitoring of the status of projects such as software development has become important. Status monitoring can be divided into two types: focused monitoring and wide area monitoring. Priority monitoring is project monitoring by the management department. Wide-area monitoring is monitoring that covers all projects and uses predetermined tools. In status monitoring, projects that are determined to have a high probability of failure or a large impact in the event of failure are targeted for priority monitoring through wide-area monitoring based on data such as costs entered into the tool. Next, with support from the management department, efforts will be made to prevent deterioration and improve projects that are subject to priority monitoring.

Conventional wide-area monitoring involves identifying projects after problems such as costs exceeding estimates become apparent. In light of the objective of leading the project to success, it is preferable to detect signs of deterioration in the project at an early stage before any problems have become apparent. However, with conventional wide-area monitoring, there is a problem in that it is difficult to detect signs of project deterioration at an early stage before problems have manifested.

Furthermore, in conventional wide-area monitoring, an investigation is conducted to determine whether or not a project of concern should really be targeted for priority monitoring. The investigation is, for example, an investigation into the scale of the project and the estimated cost excess. However, such investigations required a large amount of human resources. For this reason, priority is given to projects that would have a large impact if the situation worsens, such as projects with large order amounts, and currently monitoring is not conducted for all projects, including small-scale projects. . In other words, conventional wide-area monitoring has had the problem of high human cost in determining which projects should be targeted for priority monitoring.

Note that Patent Document 1 discloses an invention that makes it possible to identify factors that lead to risks in software development.

Japanese Patent Application Publication No. 2019-148874

In view of these circumstances, the present invention aims to detect signs of deterioration of a project at an early stage through wide-area monitoring, and to reduce the human cost of determining whether or not to target a project for priority monitoring. do.

The present invention for solving the above problems includes:
a generation unit that generates a predictive model trained using the monitoring information of the completed first project;
Input the explanatory variables of the second project in progress into the prediction model , use multiple decision trees that combine judgment conditions using the explanatory variables, and take a majority vote to determine whether the estimated cost of the second project exceeds the estimated cost. An information processing unit that outputs a predicted value and also outputs an explanatory variable that is the basis of the predicted value according to the contribution rate of the explanatory variable to the predicted value, which is determined using the SHAP (Shapley Additive exPlanations) algorithm. It is a device.

Moreover, the present invention
The information processing device
a step of generating a predictive model trained with the monitoring information of the completed first project;
Input the explanatory variables of the second project in progress into the prediction model , use multiple decision trees that combine judgment conditions using the explanatory variables, and take a majority vote to determine whether the estimated cost of the second project exceeds the estimated cost. An information processing method comprising: outputting a predicted value; and outputting an explanatory variable that is a basis for the predicted value according to a contribution rate of the explanatory variable to the predicted value, which is determined using the SHAP algorithm. It is.

According to the present invention, it is possible to detect signs of deterioration of a project at an early stage through wide-area monitoring, and to reduce the human cost of determining whether or not to target a project for intensive monitoring.

It is an example of a functional block diagram of an information processing device in this embodiment. It is an example of a screen of output information of a prediction part. It is an example of a flowchart showing processing of this embodiment.

≪First embodiment≫
[composition]
The information processing apparatus 100 shown in FIG. 1 is a computer that detects signs of deterioration in a project in progress through wide-area monitoring. The information processing device 100 includes hardware such as an input section, an output section, a control section, and a storage section. For example, when the control section is composed of a CPU (Central Processing Unit), information processing by a computer including the control section is realized by program execution processing by the CPU. Further, a storage unit included in the computer stores various programs for realizing the functions of the computer according to instructions from the CPU. This allows software and hardware to work together. The program can be provided by being recorded on a recording medium or via a network. The output unit may include the function of a display unit that displays a screen.

A project is a task to achieve a predetermined purpose. Projects can be classified into past projects that have been completed and projects that are still in progress. In this embodiment, a past project is called a "first project" and a project in progress is called a "second project." Furthermore, when past projects and in-progress projects are not distinguished, they are simply referred to as "projects."

As shown in FIG. 1, the information processing device 100 includes a generation section 1 and a prediction section 2. The information processing device 100 also stores a first project DB3 and a second project DB4.

The generation unit 1 generates a prediction model for predicting the actual estimated cost of the second project using machine learning. The actual estimated cost is an estimate of the cost that will be incurred by the end of the second project. Note that there are multiple types of costs, such as manufacturing costs and sales costs, but in this embodiment, cost refers to the economic resources consumed to achieve a specific purpose, measured in monetary terms, and manufacturing costs. It is explained as a term that includes cost of goods and cost of goods sold.
The prediction unit 2 uses the prediction model generated by the generation unit 1 to predict the actual estimated cost of the second project.
The first project DB3 is a database that stores monitoring information of the first project for each first project.
The second project DB4 is a database that stores monitoring information of the second project for each second project.

<Project monitoring information>
Project monitoring information is information for monitoring the status of a project. The monitoring information for the first project can be composed of, for example, production number information, explanatory variables, and objective variables.

The production number information is information that identifies the first project. For example, the production number information includes, but is not limited to, the production number, the progress rate (%), and the production number name.
The production number is an identifier of the first project, and can be expressed, for example, as a string of alphanumeric characters.
The progress rate is a parameter that quantitatively indicates the progress of the first project.
The production number name is the name of the first project, and can be expressed, for example, in conceivable words.

The explanatory variable is a variable that expresses the state of the first project. There are multiple types of explanatory variables. Examples of explanatory variables include, but are not limited to, work start time, work end time, actual cost, estimated cost, estimated cost, person in charge, work time, and actual man-hours. For example, explanatory variable candidates can be extracted using Boruta, and the optimal explanatory variables can be selected from among them, but the method for selecting explanatory variables is not limited to this.
The work start time is the start time (year, month, and day) of the first project.
The work end time is the end time (year, month, and day) of the first project.
Actual costs are costs incurred from the start of work to a predetermined time.
Forecast costs are costs that are expected to be incurred from a predetermined period until the end of the work.
Estimated costs are costs that are expected to be incurred from the start of work to the end of work.
The person in charge is the person(s) in charge of the first project.
The work time is the time that each person in charge spent working on the first project between the work start time and a predetermined time.
The actual number of man-hours is the number of man-hours completed between the work start time and a predetermined time, out of all the man-hours constituting the first project.
Note that the predetermined time is any time between the work start time and the work end time.

The objective variable is a variable that depends on the explanatory variable. The target variable can be, for example, the final actual cost, which is the actual cost at the end of the first project, but is not limited thereto.

(progress rate)
The progress rate can be calculated on a temporal basis, for example. For example, if the period from the start of work to the end of a project is 30 days, and the target period is the 15th day from the start of work, the progress rate will be 50%. Since the first project is a completed past project, the current progress rate is 100%. Here, the explanatory variable of the first project can be a value that changes depending on the progress rate. The monitoring information for the first project can be configured as a set of explanatory variables for each progress rate.

For example, the actual cost as an explanatory variable is a set of actual costs at all values from 0% to 100% progress rate. Actual costs at a progress rate of X% are costs incurred from the start of work to the time corresponding to a progress rate of X%. In addition, the projected cost as an explanatory variable is a set of projected costs at all values from 0% to 100% progress rate. The projected cost at a progress rate of X% is the cost that is expected to be incurred between the time corresponding to the progress rate of X% and the time at which the work is completed. Note that there are explanatory variables that do not change depending on the progress rate, such as work start time and work end time, but it is preferable to treat such explanatory variables as constants that take the same value depending on the progress rate. .

On the other hand, the monitoring information for the second project can be composed of, for example, production number information and explanatory variables.

The production number information is information that identifies the second project. For example, the production number information includes, but is not limited to, the production number, the progress rate (%), and the production number name.
The production number is an identifier of the second project, and can be expressed, for example, as a string of letters and numbers.
The progress rate is a parameter that quantitatively indicates the progress of the second project.
The production number name is the name of the second project, and can be expressed, for example, in conceivable words.

The explanatory variable is a variable that expresses the state of the second project. There are multiple types of explanatory variables. Examples of explanatory variables include, but are not limited to, work start time, work end time, actual cost, estimated cost, estimated cost, person in charge, work time, and actual man-hours. For example, explanatory variable candidates can be extracted using Boruta, and the optimal explanatory variables can be selected from among them, but the method for selecting explanatory variables is not limited to this.
The work start time is the start time (year, month, and day) of the second project.
The work end time is the end time (year, month, and day) of the second project.
Actual costs are costs incurred from the start of work to a predetermined time.
Forecast costs are costs that are expected to be incurred from a predetermined period until the end of the work.
Estimated costs are costs that are expected to be incurred from the start of work to the end of work.
The person in charge is the person(s) in charge of the second project.
The work time is the time that each person in charge spent working on the second project from the start of work to a predetermined time.
The actual number of man-hours is the number of man-hours completed between the work start time and a predetermined time, out of all the man-hours constituting the second project.
Note that the predetermined time is the current time between the work start time and the work end time.
Moreover, the actual estimated cost of the second project, which has already been explained, is the sum of the actual cost and the estimated cost of the second project.

The progress rate of the second project can be calculated as the ratio of the period from the work start time to the work end time and the period from the work start time to the present. Actual costs at a progress rate of X% corresponding to the current time are costs incurred from the start of work to the present. Further, the projected cost at a progress rate of X% corresponding to the current cost is the cost that is expected to be incurred between the current time and the end of the work. Note that there are explanatory variables that do not change depending on the progress rate, such as work start time and work end time, but it is preferable to treat such explanatory variables as constants that take the same value depending on the progress rate. .

<Prediction model>
(training)
For example, the generation unit 1 can generate a prediction model by combining a plurality of decision trees used in a random forest. Random forest is a machine learning algorithm, and is an ensemble learning algorithm that uses multiple decision trees and makes predictions by taking majority vote. The decision tree can be configured, for example, as a tree-like logic that combines judgment conditions using explanatory variables. The determination condition can be designed as appropriate, and may be, for example, whether the average daily working time of the person in charge is 5 hours or more.

The generation unit 1 can train a prediction model using the monitoring information of the first project as training data. For example, for each first project, explanatory variables at a time when the progress rate corresponds to 50% among the monitoring information of the first project can be input to the prediction model. Further, a value based on the objective variable of the monitoring information of the first project can be used as the output of the prediction model. For example, the estimated cost excess value obtained by subtracting the estimated cost from the final actual cost of the first project can be used as the output of the prediction model. The generation unit 1 trains a prediction model using monitoring information of a predetermined number of first projects.

Here, the generation unit 1 can prepare a plurality of predictive models and output the predictive models in multiple stages. For example, the generation unit 1 prepares a first prediction model and a second prediction model. The input of the first prediction model can be the monitoring information of all the first projects, and the explanatory variables at the time when the progress rate of the monitoring information is equivalent to 50% can be used. Further, the output of the first prediction model can be determined as whether or not there is an estimated cost excess (greater than 0M (0 yen)) obtained by subtracting the estimated cost from the final actual cost of the first project. Next, the input of the second prediction model is the monitoring information of the first project in which the estimated cost exceeded in the output of the first prediction model, and the explanatory variables of the monitoring information at the time when the progress rate is 50%. It can be done. Further, the output of the second prediction model can be whether the estimated cost exceeds 1 million yen (1 million yen) or more. As a result, the output of the prediction model can be classified into three values: estimated cost exceedance of 0M or less, greater than 0M and less than 1M, and 1M or more.

(prediction)
The prediction unit 2 uses the trained prediction model to predict the actual estimated cost of the second project to be predicted. For example, the prediction unit 2 inputs explanatory variables from the monitoring information of the second project at the current time, that is, at a time corresponding to a predetermined progress rate (preferably 50% or more, but may be less than 50%) into the prediction model. . Then, the prediction unit 2 can obtain a value based on the actual estimated cost as the output of the prediction model. For example, the prediction unit 2 can obtain the value of estimated cost estimate excess, which is obtained by subtracting the estimated cost from the actual estimated cost.

When the prediction model is the above-described first prediction model or second prediction model, the prediction unit 2 inputs the current explanatory variables from the monitoring information of the second project to be predicted into the first prediction model. Then, the prediction unit 2 can obtain, as an output of the first prediction model, a value indicating whether there is an excess of estimated cost (greater than 0M (0 yen)). If the estimated cost exceeds the estimated cost, the prediction unit 2 inputs the current explanatory variables from the monitoring information of the second project into the second prediction model. Then, the prediction unit 2 can obtain, as an output of the second prediction model, a value indicating whether the estimated cost estimate exceeds 1M (one million yen) or more. As a result, the estimated cost estimate excess can be classified into three values: 0M or less, greater than 0M and less than 1M, and 1M or more.
Note that 1M is an example and may be a value larger than 1M or a value smaller than 1M.

The prediction unit 2 can output information including the output of the prediction model. For example, the display unit of the information processing device 100 can display the output information of the prediction unit 2 as shown in FIG. 2 on the screen. As shown in Figure 2, the output information of the prediction unit 2 is in a tabular format with "prediction results", "crop number information", and "explanatory variables and characteristics of the prediction results" as columns and the second project as rows. It can be done.

“Prediction result” indicates the output content of the prediction model. “Prediction result” is composed of “item number”, “prediction value”, and “confidence level”.
“Item number” is a line number assigned to each second project.
The "predicted value" is the result according to the three-value classification of estimated cost estimation excess when the predictive model is a combination of the first predictive model and the second predictive model described above. "1: Exceeding 1 million yen or more" corresponds to 1 million yen or more. “2: Exceeding less than 1 million yen” corresponds to a value greater than 0M and less than 1M. “3: No problem” corresponds to 0M or less.
"Confidence" is the reliability of prediction and is expressed from 0% to 100%. For example, the confidence level can be determined using bagging, but is not limited thereto.

The "production number information" is the same as the production number information in the monitoring information of the second project.
"Explanatory variables and features of prediction results" indicate explanatory variables that contribute to the "prediction results." “Explanatory variables and features of prediction results” consists of “list of explanatory variables” and “feature ranking”.
The "list of explanatory variables" is the same as the explanatory variables of the monitoring information of the second project.
"Feature ranking" indicates the ranking of explanatory variables that contribute to the "prediction result." The higher the rank, the greater the contribution of that explanatory variable to the predicted value output. For example, the contribution rate of each variable can be determined using the SHAP (Shapley Additive exPlanations) algorithm, but is not limited thereto.

According to the first embodiment, it is possible to extract a second project in which the estimated cost exceeds 1M or more. Therefore, signs of deterioration in the project can be detected early through wide-area monitoring.

(Decision on whether or not to be subject to intensive monitoring)
The management department that has learned the output information in FIG. 2 determines whether or not to target the second project in which the estimated cost exceeds 1M or more to be subject to priority monitoring. Previously, even if AI detected signs of deterioration in a project, the basis for the AI's prediction results was a black box. For this reason, it is difficult for management departments to track down factors that indicate signs of project deterioration based on AI prediction results, and it takes a great deal of human cost to decide whether or not to target them for priority monitoring. was.

It can be said that the "feature ranking" in FIG. 2 presents the basis for why the estimated cost exceeds 1M or more. The management department can easily identify explanatory variables that greatly contribute to the prediction that the estimated cost exceeds 1M or more by referring to the "feature ranking." As a result, it becomes easier to determine whether or not to target the second project for priority monitoring, and it is possible to reduce the human cost for determining whether or not to target the second project for priority monitoring.

[process]
The processing executed by the information processing apparatus 100 is as shown in FIG. That is, first, the generation unit 1 generates a prediction model (step S1). Next, the generation unit 1 trains a prediction model using the monitoring information of the first project at a predetermined progress rate (step S2). Next, the prediction unit 2 uses the prediction model to output a predicted value of the estimated cost estimate excess of the target second project and an explanatory variable that contributes to the predicted value as a basis for prediction (step S3). The management department determines whether or not to target the second project, which shows signs of deterioration, for priority monitoring based on the basis of the prediction.

≪Second embodiment≫
When describing the second embodiment, points that are different from the first embodiment will be explained, and explanations of overlapping points will be omitted. In the first embodiment, the explanatory variables of the monitoring information of the first project serving as training data are explanatory variables at a time when the progress rate is equivalent to 50%. In the second embodiment, the timing of explanatory variables of the first project used for training data is regularized.

For example, if the period of the first project, that is, the period from the start of work to the end of work, is approximately several months, the time of the explanatory variable used for training data, that is, the training date (learning day), is set to the 25th of every month. do.
Therefore, if the work start time of the first project of crop number A is 4/15 and the work end time is 6/30, the training dates of the first project of crop number A are 4/25 and 5/25. becomes. In other words, among the monitoring information of the first project of production number A, the explanatory variable at 4/25 (an explanatory variable at a progress rate equivalent to 4/25) and the explanatory variable at 5/25 (an explanatory variable at a progress rate equivalent to 5/25) A total of two explanatory variables (explanatory variables) are input to the prediction model.
In addition, if the work start time of the first project of crop number B is 5/1 and the work end time is 9/15, the training dates of the first project of crop number B are 5/25 and 6/25. , 7/25, 8/25. In other words, among the monitoring information of the first project of production number B, the explanatory variable at 5/25 (an explanatory variable at a progress rate equivalent to 5/25) and the explanatory variable at 6/25 (an explanatory variable at a progress rate equivalent to 6/25) explanatory variable at 7/25 (explanatory variable at progress rate equivalent to 7/25), explanatory variable at 8/25 (explanatory variable at progress rate equivalent to 8/25) A total of four times are used as input for the prediction model.
Further, if the work start time of the first project of production number C is 6/1 and the work end time is July 10, the training date of the first project of production number C is 6/25. That is, a total of one explanatory variable at 6/25 (an explanatory variable at a progress rate equivalent to 6/25) of the monitoring information for the first project of production number C is input to the prediction model.

As a result, for most of the first project, explanatory variables with multiple types of progress rates are input into the prediction model. Using the prediction model trained in this way, the prediction unit 2 predicts the actual estimated cost of the second project. In this case, even if the progress rate of the second project is a low progress rate (for example, about 30%) and explanatory variables at a time corresponding to the low progress rate are input into the prediction model, the predicted value output by the prediction unit 2 It was confirmed that the confidence level (see Figure 2) was sufficiently high.

According to the second embodiment, training days can be regularized, and explanatory variables at multiple types of progress rates can be input into the prediction model for the same first project. As a result, the actual estimated cost of the second project can be predicted earlier.
Furthermore, by regularizing the training dates, input to the prediction model for all the first projects can be systematized, and the processing required for training can be simplified.

[Modified example]
(a): In the first and second embodiments, the progress rate was calculated on a timing basis using the project period. However, for example, the progress rate may be calculated from the degree of accomplishment of the work undertaken in the project.
(b): In the first embodiment, for each first project, explanatory variables at a time corresponding to a progress rate of 50% among the monitoring information of the first project are input to the prediction model. However, for example, for each first project, explanatory variables at any period corresponding to the same progress rate other than 50% may be input to the prediction model. Furthermore, explanatory variables at times corresponding to different progress rates for each first project may be input into the prediction model.
(c): In the second embodiment, by regularizing the training days, multiple types of progress rates can be effectively selected for the same first project, and the explanatory variables at the selected progress rates can be used as predictive models. entered. However, for example, the user of the information processing device 100 operates the input unit to select multiple types of arbitrary progress rates for the same first project, and inputs explanatory variables at the selected progress rates into the prediction model. It's okay.

(d): It is also possible to realize a technique that combines the various techniques described in this embodiment as appropriate.
(e): The software described in this embodiment can be implemented as hardware, and the hardware can also be implemented as software.
(f): Other changes may be made as appropriate to the hardware, software, flowcharts, etc. without departing from the spirit of the present invention.

100 Information processing device 1 Generation unit 2 Prediction unit 3 First project DB
4 2nd project DB

Claims (6)

a generation unit that generates a predictive model trained using the monitoring information of the completed first project;
Input the explanatory variables of the second project in progress into the prediction model , use multiple decision trees that combine judgment conditions using the explanatory variables, and take a majority vote to determine whether the estimated cost of the second project exceeds the estimated cost. An information processing unit that outputs a predicted value and also outputs an explanatory variable that is the basis of the predicted value according to the contribution rate of the explanatory variable to the predicted value, which is determined using the SHAP (Shapley Additive exPlanations) algorithm. Device. The prediction model inputs a first prediction model that outputs whether or not there is an excess of the estimated cost, and an explanatory variable of the second project for which the first prediction model has an excess of the estimated cost, and The information processing device according to claim 1, which is combined with a second prediction model that outputs whether the estimated cost estimate exceeds a predetermined value or not. 3. The information processing apparatus according to claim 1, wherein the monitoring information of the first project used for training the prediction model is monitoring information of the first project at a predetermined progress rate. The information processing device according to claim 3, wherein a plurality of said predetermined progress rates are set. 5. The information processing apparatus according to claim 4, wherein when the predetermined progress rate is determined based on a timing standard with respect to the period of the first project, the time corresponding to the predetermined progress rate is regularized. The information processing device
a step of generating a predictive model trained with the monitoring information of the completed first project;
Input the explanatory variables of the second project in progress into the prediction model , use multiple decision trees that combine judgment conditions using the explanatory variables, and take a majority vote to determine whether the estimated cost of the second project exceeds the estimated cost. An information processing method comprising: outputting a predicted value; and outputting an explanatory variable that is a basis for the predicted value according to a contribution rate of the explanatory variable to the predicted value, which is determined using the SHAP algorithm. .

Images