High-precision classification condition generation device using decision trees and high-precision classification condition generation method using decision trees
The decision tree classification condition generation device enhances classification accuracy by regenerating trees to incorporate missed conditions, addressing the issue of reduced precision in existing models.
Patent Information
- Application Number
- JP2023083834
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-05-22
- Publication Date
- 2024-12-04
AI Technical Summary
Existing decision tree models suffer from poor classification accuracy due to the influence of initial branching conditions, leading to missed effective classification conditions and reduced precision.
A decision tree classification condition generation device that includes a decision tree generation information collection unit, a decision tree generation section, a decision tree re-creation section, and an output section to generate and regenerate decision trees based on collected information, ensuring highly accurate branching conditions.
The solution enables the determination of highly accurate branching conditions in decision trees, improving classification precision by regenerating trees to include missed effective conditions.
Smart Images

Figure 2024167625000001_ABST
Abstract
Description
[Technical field]
[0001] The present invention relates to an apparatus for generating high-precision classification conditions using a decision tree and a method for generating high-precision classification conditions using a decision tree. [Background technology]
[0002] In failure prediction and anomaly detection, decision trees are tree-structured graphs generated by repeatedly dividing features based on conditions. They are also known as machine learning analysis using auto-generation algorithms and are used to automatically generate explainable detection conditions based on collected data using auto-generation algorithms.
[0003] Patent Document 1 discloses a method for explaining a predictive model, in which a computer executes a selection process, a creation process, an identification process, a re-creation process, and an output process. The selection process selects a plurality of models that can linearly separate data included in a dataset based on a dataset input to the predictive model and a prediction result of the predictive model for the dataset.
[0004] The creation process creates a decision tree in which each of the selected models is a leaf and each logic for classifying data included in the dataset is a node. The identification process identifies branches to be pruned in the decision tree based on the variability of data belonging to the leaves of the created decision tree.
[0005] The re-creation process re-creates the decision tree based on a data set corresponding to the decision tree with the identified branches pruned, and the output process outputs logic corresponding to each node of the re-created decision tree as an explanatory result of the prediction model. [Prior art documents] [Patent documents]
[0006] [Patent Document 1] Patent Publication No. 2021-71823 Summary of the Invention [Problem to be solved by the invention]
[0007] In Patent Document 1, data that reduces classification accuracy is removed, and a decision tree is generated using a general generation algorithm that generates a model with high classification accuracy. The data variance (poor classification accuracy) of each node of the decision tree is calculated, the node with the greatest variance is identified, and the decision tree is regenerated by excluding the data belonging to the identified node, and the decision tree with variance below a predetermined value is adopted as the model.
[0008] In a decision tree, due to the influence of previously generated branching conditions, other branching conditions may not be generated, and some conditions effective for classification may not be obtained, resulting in a decrease in accuracy. [Means for solving the problem]
[0009] The object of the present invention is achieved by a classification condition generation device for a decision tree, which comprises a decision tree creation information collection unit that collects decision tree creation information and creates an information table for decision tree generation, a decision tree generation unit that generates a decision tree from the information table for decision tree generation, a decision tree re-creation unit that re-creates a decision tree when classification conditions can be created from the generated decision tree, and an output unit that outputs the created classification conditions to the classification condition table. Effect of the Invention
[0010] According to the present invention, it is possible to obtain a branching condition of a decision tree with high accuracy. [Brief description of the drawings]
[0011] [Figure 1] Example of classification condition generation system according to the embodiment of the present invention [Diagram 2] Example of an error log table in an embodiment of the present invention [Diagram 3] Example of a fault record table in an embodiment of the present invention [Figure 4] 1 is a flowchart showing a classification condition generation process according to an embodiment of the present invention; [Diagram 5] Example of decision tree generation information table in the embodiment of the present invention [Figure 6] 1 is a flowchart showing a process for generating a decision tree according to an embodiment of the present invention; [Figure 7] An example of a decision tree generated by the decision tree generation process in the embodiment of the present invention. [Figure 8] An example of a diagram showing the first classification condition of a decision tree generated by the decision tree generation process in an embodiment of the present invention, using a decision tree generation information table. [Figure 9] An example of a diagram showing the second classification condition of a decision tree generated by the decision tree generation process in an embodiment of the present invention using a decision tree generation information table. [Figure 10] An example of a diagram showing a third classification condition of a decision tree generated by a decision tree generation process in an embodiment of the present invention, using a decision tree generation information table. [Figure 11] Example of a classification condition table in the embodiment of the present invention [Figure 12] An example of a diagram showing the basis for generating classification conditions in a decision tree generation information table in an embodiment of the present invention. [Figure 13] 1 is a flowchart illustrating a decision tree regeneration process according to an embodiment of the present invention. [Figure 14] 1 is an example of a decision tree generation information table with reduced feature amounts used to generate a first regenerated decision tree in the decision tree regeneration process in an embodiment of the present invention. [Figure 15] An example of a first regenerated decision tree generated by a decision tree regeneration process in an embodiment of the present invention. [Figure 16] An example of a classification condition table to which classification conditions obtained from the first regenerated decision tree have been added in an embodiment of the present invention. [Figure 17] An example of a diagram showing a first classification condition of a first regenerated decision tree generated in a decision tree regeneration process in an embodiment of the present invention, using a decision tree generation information table. [Figure 18] An example of a diagram showing the second classification condition of the first regenerated decision tree generated in the decision tree regeneration process in an embodiment of the present invention, using a decision tree generation information table. [Figure 19]An example of a decision tree generation information table with reduced feature amounts used to generate a second regenerated decision tree in the decision tree regeneration process in an embodiment of the present invention. [Figure 20] An example of a second redecision tree generated by the decision tree regeneration process in an embodiment of the present invention. [Figure 21] An example of a diagram showing a first classification condition of a second regenerated decision tree generated in a decision tree regeneration process in an embodiment of the present invention, using a decision tree generation information table. [Figure 22] Example of a classification condition display screen in the embodiment of the present invention DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0012] Hereinafter, embodiments of the present invention will be described with reference to the drawings. In each drawing for explaining the embodiments, the same components are given the same names and reference numerals as much as possible, and repeated description will be omitted.
[0013] The present invention is not limited to the following examples, and includes various modifications and equivalent configurations within the spirit of the appended claims. For example, the above-mentioned examples have been described in detail to clearly explain the present invention, and the present invention is not necessarily limited to those including all of the described configurations.
[0014] In addition, the processing units described in the embodiments may be realized in hardware, for example by designing some or all of them as integrated circuits, or may be realized in software by a processor interpreting and executing a program that realizes each function.
[0015] The tables, areas, etc. described in the embodiments may be a database (DB) or may be data stored in the main memory. EXAMPLES
[0016] 1 is an example of a block diagram showing an example of the configuration of a classification condition generation system in an embodiment of the present invention. The classification condition generation system is composed of a classification condition generation device 100, a monitored system 120, and an input / output device 109. The monitored system 120 in the embodiment is configured to include a plurality of HDDs. The monitored objects of the invention are not limited to HDDs, and can also be used for detecting signs of failure and abnormalities in devices, plants, information processing systems, etc.
[0017] In this embodiment, for ease of understanding, the explanation will be given using the example of HDD failure sign detection and anomaly detection, but the present invention is applicable not only to binary classification such as anomaly detection and failure sign detection, but also to devices and methods that perform classification based on two or more conditions in various fields such as marketing and medicine.
[0018] The classification condition generating device 100 and the monitored system 120 are connected via a network interface 110 to a wired or wireless communication network 111 such as the Internet, a LAN, a WAN, or a dedicated line.
[0019] In this example, it is assumed that the classification condition generation device 100 and the monitored system 120 are connected to the same communication network 111, but the classification condition generation device 100 may be connected to a network different from that of the monitored system. For example, all or part of the program of the classification condition generation device 100 may be realized by a service provided by a cloud system via an API (Application Programming Interface) or the like.
[0020] The input / output device 109 is composed of output devices such as a display and a touch panel, and input devices such as a keyboard, a mouse, and a touch panel. The processing device 101 is composed of a CPU (Central Processing Unit), the memory 102 is composed of a RAM (Random Access Memory), a ROM (Read Only Memory), etc., the external storage device 103 is composed of a HDD (Hard Disc Drive) or an SSD (Solid State Drive), etc., and the network interface 110 is composed of a NIC (Network Interface Card), a wireless communication module, a USB (Universal Serial Interface) module, a serial communication module, etc.
[0021] The classification condition generation device 100 includes processing units including an information collection unit 104, a classification condition generation unit 105, a decision tree generation unit 106, a classification condition extraction unit 107, and a decision tree regeneration unit 108, as well as tables including an error log table 200, a fault record table 300, an information table for decision tree generation 400, and a classification condition table 500 in an external storage device 103.
[0022] These processing units are configured as software modules, which are loaded into memory 102 by CPU 101 and executed by referring to tables stored in external storage device 103 .
[0023] The information collecting unit 104 is a software module that collects information on the monitored system. In this example, the information is collected by the classification condition generating device 100, but the information may be collected in the monitored system and sent to the classification condition generating device 100.
[0024] The classification conditions generated by the classification condition generating device 100 are displayed to the user by the input / output device 109, and a selection as to whether or not to use the classification conditions is accepted from the user.
[0025] 2 is an example of an error log table in an embodiment of the present invention. The error log table 200 is information that records errors that have occurred in HDDs that constitute the monitored system 120, and includes an error occurrence date and time 201, an HDD identifier 202 of the HDD in which the error occurred, and an error content 203 that indicates the type of error that has occurred.
[0026] 3 is an example of a failure record table in an embodiment of the present invention. The failure record table is information that records failure cases of HDDs that constitute a monitored system, and includes a failure date and time 301 that indicates the date and time of the failure, and an HDD identifier 302 of the failed HDD.
[0027] 4 is an example of a flowchart for explaining an outline of a classification condition generation process according to an embodiment of the present invention. The classification condition generation process is executed, for example, when a predetermined input is made to the classification condition generation device 100 by a user, or at a predetermined timing (for example, at a predetermined time or at a predetermined time interval).
[0028] First, the information collecting unit 104 refers to the information in the error log table 200 and the fault record table 300, and generates a decision tree generation information table 400 (S10). The decision tree generating unit 106 generates a decision tree using the generated decision tree generation information table 400 (S100).
[0029] Once the classification conditions have been generated from the decision tree (S20), the decision tree regeneration unit 108 regenerates the decision tree (S200). The classification conditions obtained in the decision tree generation process S100 and the decision tree regeneration process S200 are registered in the classification condition table 500, and the classification condition table 500 is output to the input / output device 109 (S30).
[0030] 5 shows an example of a decision tree generation information table in an embodiment of the present invention. The decision tree generation information table 400 is a table generated in S10 of FIG.
[0031] It is composed of error information 401 indicating the type and number of errors that have occurred, and a classification 402 indicating whether the HDD has subsequently failed. In this example, if there is a failure, the classification is abnormal, and if there is no failure, the classification is normal. In this example, classification into two categories, normal and abnormal, is explained as an example, but the present invention is also applicable to classification into three or more categories.
[0032] In the case of a failed HDD, errors that occurred within a certain period (for example, two weeks) prior to the date of failure are referenced in the error log table 200. In this example, the number of errors that occurred is collectively counted by day.
[0033] For example, in the case of HDD-001, which was confirmed to have failed on 1 / 10 / 2023 in failure record table 300, first, by looking up errors that occurred within a certain period (e.g., two weeks) from the date of the failure in error log table 200, it is found that the first, second, and fourth rows of error log table 200 are errors in HDD-001.
[0034] According to these three error logs, three instances of error A occurred in HDD-001 on 2023 / 1 / 5, which corresponds to the first row of the decision tree generation information table 400. Similarly, for HDD-003, which failed on 2023 / 1 / 14, referring to the sixth and seventh rows in the error log table 200, two instances of error A occurred on 2023 / 1 / 8, which corresponds to the second row of the decision tree generation information table 400.
[0035] On the other hand, the 7th, 8th, and 9th rows in the decision tree generation information table 400, which are classified as normal, represent information about errors that occurred in a non-faulty HDD, and in this example, the information is not limited to errors that occurred within a certain period of time (e.g., two weeks), such as when the HDD is not faulty or has a fault.
[0036] In this embodiment, conditions for classifying the information in this decision tree generating information table 400 into classifications 402 are automatically generated based on error information 401. That is, the number of errors A, B, and C is used as a hint to divide the table into item numbers 1 to 6 (abnormal) and 7 to 8 (normal). This classification condition based on the number of errors A, B, and C means a condition for determining whether or not the HDD will fail later based on the number of errors. Therefore, the classification condition generated in this embodiment can be used as a condition (threshold condition) for detecting a sign of HDD failure.
[0037] For example, referring to 401a showing the number of occurrences of error A, all rows with a value of 2 or more are classified as abnormal. This indicates that if error A occurs more than twice in one day, the HDD will subsequently fail. Similarly, classification conditions for errors B and C are also generated, with the objective of generating conditions that allow classification according to classification 402a or close to classification 402a.
[0038] 6 is an example of a flowchart illustrating a decision tree generation process according to an embodiment of the present invention. After the decision tree generation information table 400 is generated, the decision tree generation unit 106 generates a decision tree from this table.
[0039] First, the number of feature amounts of error information in the decision tree generation information table 400 is referenced, and this value is set as N (S101). Next, a decision tree of height N (the number of types of error information) is generated by an automatic generation algorithm, with the error information 401 in the decision tree generation information table as the explanatory variable and the classification 402 as the objective variable (S102). In the example shown in Fig. 5, there are three types of feature amounts, namely error A (401a), error B (401b), and error C (401c), and N=3.
[0040] If all the nodes of the decision tree are normal or not abnormal (S103), the branches to the abnormal node are traced and a branch condition is extracted (S104). The extracted branch condition is then stored in the classification condition table 500 (S105). If all the nodes of the decision tree are normal or all abnormal, the process ends.
[0041] FIG. 7 shows an example of a decision tree generated by the decision tree generation process in the embodiment of the present invention. The decision tree is generated by any automatic decision tree generation algorithm. The decision tree performs classification based on the type of error specified in the explanatory variable so as to be the same as or close to the classification specified in the objective variable. 7, if the condition of each branch is satisfied, the result is classified into the left node, and if not, the result is classified into the right node. For the sake of explanation, 601a is the first classification condition of the decision tree, 601b is the second classification condition, and 601c is the third classification condition.
[0042] The terminal nodes of the decision tree are the classification results, with nodes with more normal results being classified as normal nodes and nodes with more abnormal results being classified as abnormal nodes.
[0043] FIG. 8 is an example of a diagram showing the first classification condition of a decision tree generated in the decision tree generation process according to the embodiment of the present invention, using a decision tree generation information table.
[0044] The first classification condition 601a of the decision tree is “error A<2”, and the information is classified into 602a which satisfies this and 602b which does not. If we explain this using the decision tree generation information table 400, it is divided into information 4012 where error A is 2 or more and other information 4013, as shown in 4011.
[0045] The classifications of the information 4012 classified as not satisfied 602b are all "abnormal" and no further classification is necessary, whereas the information 4013 classified as satisfied 602a includes "abnormal" and "normal".
[0046] FIG. 9 is an example of a diagram showing the second classification condition of a decision tree generated in the decision tree generation process according to the embodiment of the present invention, using a decision tree generation information table.
[0047] In Fig. 8, 4013 is then further classified by a second classification condition 601b of the decision tree. Here, the second classification condition 601b does not relate to 4012, since it classifies only 4013 in Fig. 8. The second classification condition 601b of the decision tree is "error B<1", and is classified into 603a that satisfies this and 603b that does not, and if this is explained using the decision tree generation information table 400, 4013 in Fig. 8 is divided into information 4022 where error B is 1 or more and other information 4023 as shown in 4021.
[0048] The classifications of the information 4023 classified as "satisfied" 603a are all "normal" and no further classification is necessary, whereas the classifications of the information 4022 classified as "not satisfied" 603b include "normal" and "abnormal".
[0049] Fig. 10 is an example of a diagram showing the third classification condition of a decision tree generated by the decision tree generation process in the embodiment of the present invention, using a decision tree generation information table. As in Fig. 9, if 4022 in Fig. 9 is classified into 4032 and 4033, it can be completely classified into "normal" and "abnormal".
[0050] 11 is an example of a classification condition table in the embodiment of the present invention. The results of extracting conditions for classifying as abnormal from the decision tree shown in FIG.
[0051] In the decision tree, information classified as abnormal follows the path 602b or 602a → 603b → 604b. When this is made into a conditional formula, 602b is "Error A ≧ 2", and 602a → 603b → 604b is "Error A < 2 and Error B ≧ 1 and Error A ≧ 1", so the condition for classification as abnormal is "Error A ≧ 2 or (Error A < 2 and Error B ≧ 1 and Error A ≧ 1)".
[0052] If we rearrange this equation according to the basic formula for logical expressions, we get Error A ≥ 2 or (Error A < 2 and Error B ≥ 1 and Error A ≥ 1) = (error A ≥ 2 or error A < 2) and (error A ≥ 2 or (error B ≥ 1 and error A ≥ 1)) = Error A ≥ 2 or (Error B ≥ 1 and Error A ≥ 1) can be converted to
[0053] Therefore, the conditions that can be extracted from the decision tree shown in FIG. 7, that is, error A≧2 and error B≧1 and error A≧1, are stored in the classification condition table 500 as 501a and 501b.
[0054] FIG. 12 is an example of a diagram showing points that serve as grounds for generating classification conditions in a decision tree generating information table according to an embodiment of the present invention. Focusing on 4041, the classification is abnormal when the value of error A is 2 or more. Therefore, we can expect to obtain the classification condition of error A ≥ 2. Focusing on 4042, if the value of error A is 1 or more and the value of error B is 1 or more, the classification is abnormal. Therefore, we can expect the classification conditions to be error A ≥ 1 and error B ≥ 1. Focusing on 4043, the classification is abnormal when the value of error B is 2 or more. Therefore, we can expect to obtain the classification condition of error B ≥ 2. Focusing on 4044, the classification is abnormal if the value of error C is 2 or more. Therefore, we can expect to obtain error C ≥ 2 as the classification condition. From the above, it is expected that error A ≥ 2, error A ≥ 1 and error B ≥ 1, error B ≥ 2, and error C ≥ 2 will be obtained, but error B ≥ 2 and error C ≥ 2 are not obtained from the decision tree shown in Fig. 7. The decision tree regeneration process S200 described below is a process for obtaining these conditions.
[0055] 13 is an example of a flowchart for explaining the decision tree regeneration process in an embodiment of the present invention. In the previous decision tree, a feature value that appears in the first generated branch is identified (S201). At this time, the previous decision tree is the decision tree shown in FIG. 7 that was generated in the decision tree generation process S100, and the first generated branch is the top first classification condition 601a, which is the condition "error A<2". Therefore, the feature value identified here is error A.
[0056] The identified feature, in this example, error A, is removed from the information for generating a decision tree (S202), and the number of remaining features is set to N (S203). Fig. 14 shows the state of the information table for generating a decision tree after error A has been removed.
[0057] If N is 0, that is, if there are no more features (S204), the decision tree cannot be regenerated and the process ends. If N>0, the decision tree is regenerated using decision tree generation information with one feature removed (S205).
[0058] If the regenerated decision tree cannot classify the data, that is, if all the nodes of the decision tree are normal or abnormal (S206), the process ends. If the nodes of the decision tree include both normal and abnormal data, classification conditions are extracted from the regenerated decision tree (S207).
[0059] If no new classification conditions are obtained from the regenerated decision tree (if all the conditions obtained from the regenerated decision tree already exist in the classification condition table 500) (S208), the process ends.
[0060] If the newly obtained classification condition does not exist in the classification condition table 500, the newly obtained classification condition is added to the classification condition table 500 (S209), and the process returns to S201.
[0061] Fig. 15 shows a regenerated decision tree generated from the decision tree generation information table of Fig. 14 with error A removed. Fig. 16 shows the classification condition table with the classification condition added.
[0062] FIG. 17 is an example of a diagram showing the first classification condition of the first regenerated decision tree generated in the decision tree regeneration process in the embodiment of the present invention, using a decision tree generation information table.
[0063] The first classification condition 701a of the first regenerated decision tree is “error B<2”, and the information is classified into 702a which satisfies this condition and 702b which does not. If we explain this using the decision tree generation information table shown in FIG. 14 with error A removed, the information is divided into information 4102 where error B is 2 or more, and other information 4103, as shown in 4101.
[0064] The classification of the information 4102 classified as 702b, which does not satisfy "error B<2", is all "abnormal" and does not require any further classification, whereas the classification of information 4103 classified as 702a, which satisfies "error B<2", includes both "abnormal" and "normal".
[0065] FIG. 18 is an example of a diagram showing the second classification condition of the first regenerated decision tree generated in the decision tree regeneration process in the embodiment of the present invention, using a decision tree generation information table.
[0066] The second classification condition 701b of the first regenerated decision tree is “error C<2”, and the information is classified into 703a which satisfies this condition and 703b which does not. If we explain this using the decision tree generation information table shown in FIG. 15 with error A removed, 4103 in FIG. 17 is divided into information 4112 where error C is 2 or more, and the other information 4113, as shown in 4111.
[0067] All of the information classified as 4112 in 703b, which does not meet the criteria for "error C<2", is classified as "abnormal" and does not require further classification, whereas information classified as 4113 in 703a, which meets the criteria for "error C<2", includes both "abnormal" and "normal".
[0068] Although 4113 has not been completely classified, it cannot be classified any further due to the values of error B and error C. 4113, which is classified as 703a in FIG. 15, has two "abnormal" cases and three "normal" cases, so it is classified as "normal."
[0069] FIG. 19 shows an example of a decision tree generation information table with reduced features used for generating a second regenerated decision tree in the decision tree regeneration process in the embodiment of the present invention.
[0070] From Figure 14, the feature of error B was further removed, leaving only error C.
[0071] 20 is an example of a second re-decision tree generated by the decision tree regeneration process in the embodiment of the present invention. Error C alone cannot be classified into "normal" and "abnormal".
[0072] 21 is an example of a diagram showing the first classification condition of the second regenerated decision tree generated by the decision tree regeneration process in the embodiment of the present invention, using a decision tree generation information table. Error C alone cannot be classified into "normal" and "abnormal".
[0073] 22 is an example of a screen that outputs the generated classification conditions in an embodiment of the present invention. The classification condition display section 910 includes a failure probability 911, which is the probability that an error for which a threshold has been set, a threshold value, and data used to generate the decision tree, that satisfies this condition and is a failure, and assists in the decision of application. In addition, a details button 912 that presents failed / operating HDDs that satisfy the generated condition and their error information is displayed for each error.
[0074] A Details All Display button 914 is displayed to display the details of multiple conditions at once, and an Apply to Detection button 915 is displayed to select the conditions to be used for actual detection from the generated conditions, and the target classification conditions are selected using check boxes 913.
[0075] When this condition is applied, the detailed information display unit 920 may output information 921 of a failed HDD that satisfies the condition, which is information of a HDD in which a symptom of failure has been correctly detected. In this case, the identifier of the HDD, the date on which the HDD was introduced, the date on which the HDD failed, the date on which an error was observed, and error information such as error A and error B are displayed.
[0076] This screen is displayed by providing an error information selection section 600 which searches the error log table 200 for a HDD in which a target error has occurred during a predetermined period, and selects from the failure record table 300 the HDD in which the error has occurred that is experiencing a failure.
[0077] When this condition is applied, information 922 of an operating HDD that satisfies the condition, which is information of a normal HDD whose failure symptom has been erroneously detected due to this condition, may be output. In this case, the identifier of the HDD, the date on which the HDD was introduced on the left, the date on which the error displayed on the right was observed, and error information such as error A and error B are displayed.
[0078] This information can be used to determine the number of false positives caused by the selected conditions, as well as to identify HDDs that are not broken but have a high number of errors.
[0079] This screen is displayed by providing an error information selection section 600 which searches the error log table 200 for a HDD in which a target error has occurred during a predetermined period, and selects from the failure record table 300 an HDD in which the error has occurred but which is not faulty. [Explanation of symbols]
[0080] 100 Classification condition generating device, 101 Processing unit (CPU), 102 Memory, 103 External storage device, 104 Information collecting unit, 105 Classification condition generating unit, 106 Decision tree generating unit, 107 Classification condition extracting unit, 108 Decision tree regenerating unit, 109 Input / output device, 110 Network interface, 111 Communication network, 120 Monitored system, 121, 122 Hard disk, 200 Error log table, 300 Fault record table, 400 Information table for decision tree generation, 500 Classification condition table, 600 Error information selecting unit
Claims
1. In the device for generating classification conditions for a decision tree, a decision tree creation information collecting unit that collects decision tree creation information and creates a decision tree generation information table; a decision tree generation unit that generates a decision tree from the decision tree generation information table; a decision tree re-creation unit that re-creates a decision tree when a classification condition can be created from the generated decision tree; The classification condition generating device for a decision tree includes an output unit that outputs the generated classification conditions to a classification condition table.
2. 2. The apparatus for generating classification conditions for a decision tree according to claim 1, A classification condition generation device for a decision tree, in which, when a decision tree is created and contains both a node that satisfies a first condition and a node that satisfies a second condition, the decision tree generation unit follows the branches to the node that satisfies the second condition, obtains the classification conditions, and adds the obtained classification conditions to a classification condition table.
3. 2. The apparatus for generating classification conditions for a decision tree according to claim 1, The decision tree re-creation unit deletes the explanatory variables of the initially generated classification conditions from the decision tree generation information table, If explanatory variables remain, a decision tree is recreated using the decision tree generation information table from which explanatory variables of the initially generated classification conditions have been deleted. When the created decision tree contains both a node that satisfies a first condition and a node that satisfies a second condition, the device follows the branches up to the node that satisfies the second condition to obtain a classification condition, and adds the obtained classification condition to a classification condition table.
4. 4. The apparatus for generating classification conditions for a decision tree according to claim 3, A classification condition generating device for a decision tree that terminates processing when a classification condition determined from a reconstructed decision tree is included in the classification condition table.
5. 2. The apparatus for generating classification conditions for a decision tree according to claim 1, The output unit is a decision tree classification condition generating device that outputs the created classification condition of the decision tree in association with the probability of occurrence of the second condition.
6. 2. The apparatus for generating classification conditions for a decision tree according to claim 1, an input section for receiving a selection of classification conditions; An information selection unit that selects decision tree creation information that satisfies the received classification condition, The output unit outputs decision tree creation information that satisfies the classification conditions of the created decision tree and the classification conditions selected by the information selection unit.
7. 2. The apparatus for generating classification conditions for a decision tree according to claim 1, an input section for receiving a selection of classification conditions; An information selection unit that selects decision tree creation information that does not satisfy the received classification condition, The output unit outputs the classification conditions of the created decision tree and decision tree creation information that does not satisfy the classification conditions selected by the information selection unit.
8. In the method for generating classification conditions for a decision tree, A decision tree creation information collection unit collects decision tree creation information and creates an information table for generating a decision tree; A decision tree generating unit generates a decision tree from the decision tree generation information table, When the decision tree re-creation unit is able to create a classification condition from the generated decision tree, the decision tree is re-created; A method for generating classification conditions for a decision tree, in which an output unit outputs the created classification conditions to a classification condition table.
9. The method for generating classification conditions for a decision tree according to claim 8, A method for generating classification conditions for a decision tree, in which, when a decision tree is created and the decision tree contains both a node that satisfies a first condition and a node that satisfies a second condition, the decision tree generation unit follows the branches up to the node that satisfies the second condition, obtains the classification conditions, and adds the obtained classification conditions to a classification condition table.
Citation Information
Patent Citations
Train control on-board equipment fault prediction method based on fault characteristic relation graph
CN114818353A
Inverse model calculation apparatus and inverse model calculation method
JP2005242979A
Data processor, data processing method and data processing program
JP2006048129A
Interactive adjustment of decision rules
US20190197141A1
Forecasting model explanation method and forecasting model explanation program and forecasting model explanation device
JP2021071823A
Cited By
Menu structure automatic generation method and system based on submission task configuration
CN120762667A
