Method for generating classification model and classification method using same

Abstract

The embodiment of the invention relates to a method for generating a classification model and a classification method using the model. A computer-implemented method for generating a classification model includes obtaining at least one set of learning data, identifying at least one characteristic of the learning data to be investigated, extracting a value for each characteristic defined for all the learning data, identifying a range of values for each characteristic from the extracted values, creating a classification table, and generating a classification model based on the classification table. A category is assigned to each cell of the classification table according to the number of occurrences of the extracted value according to each characteristic of each set of learning data with respect to a range of values defined for each study characteristic, and a classification model including the classification table is generated.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims the benefit of French Patent Application No. 2305829, filed on June 9, 2023, which is incorporated herein by reference. Technical Field

[0003] The present invention relates generally to learning data, and in particular embodiments to methods for generating classification models. Background Art

[0004] Computers can use classification models to classify or group data according to some characteristics. There are several types of classification models. Each classification model type has its advantages and disadvantages depending on the type of data to be classified.

[0005] In particular, classification models include decision trees, neural networks, neighbor-based classification models such as k-nearest neighbor (KNN), and support vector machines (SVM).

[0006] Furthermore, a microcontroller is a programmable integrated circuit that can be used to implement a classification model.

[0007] However, microcontrollers have limitations in terms of memory and computing power resources. In fact, microcontrollers usually have limited memory storage capacity. Microcontrollers also have limited processing speed.

[0008] These limitations may restrict their ability to implement some classification models.

[0009] The size and complexity of the classification model and the available resources on the microcontroller should be considered to determine whether it is possible and desirable to use the classification model on the microcontroller.

[0010] In particular, classification models that require large amounts of data storage may exceed the storage capacity of a microcontroller. In addition, classification algorithms that require large amounts of computation may take a long time to execute on a microcontroller.

[0011] Therefore, some classification models are not suitable for use with microcontrollers. In general, microcontrollers are best suited for simple classification models.

[0012] It is therefore desirable to provide a solution that allows classification to be implemented in a simple, fast and cheap way in terms of memory footprint. Summary of the invention

[0013] According to one embodiment, a computer-implemented method for generating a classification model is provided, the method comprising:

[0014] - obtaining at least one set of learning data, each set of learning data being associated with an indicated category,

[0015] - identifying at least one characteristic of the learning data to be studied,

[0016] - extract the value of each feature found for all learning data,

[0017] - set the range of values ​​for each property from the extracted values,

[0018] - creating a classification table having a number of dimensions corresponding to the number of studied characteristics, each dimension having a size equal to the number of ranges of values ​​set for the characteristic associated with that dimension, each cell of the classification table being associated with a range of values ​​for each studied characteristic,

[0019] - assigning a class to each cell of the classification table according to the number of occurrences of the extracted values ​​of each characteristic of each set of learning data relative to the range of values ​​set for each characteristic studied,

[0020] -Generate a classification model including a classification table.

[0021] Such a generation method allows obtaining a classification model that is simple and fast to use and occupies relatively reduced space in the memory.

[0022] Thus, such a classification model can be used by a computer system comprising limited resources in terms of memory and computing power. In particular, such a classification model can be used by a microcontroller.

[0023] Since categories are assigned to the range of values ​​of each studied characteristic, the classification model is set to improve its generalization to data not used as learning data while avoiding overfitting the learning data. The limited number of value ranges related to the number of learning data also allows avoiding overfitting.

[0024] Such classification models can also be easily updated by taking into account new learning data.

[0025] Furthermore, if the learning data is divided into several groups of different sizes, weights can be applied to the number of occurrences of the learning data used when assigning categories to different cells of the classification table. Therefore, such a classification model generation method is applicable to unbalanced sets of learning data.

[0026] In an advantageous implementation, the classification model also comprises minimum and maximum values ​​for each investigated characteristic.Such a classification model can be stored in a relatively reduced space in a memory.

[0027] Alternatively, the classification model may include all ranges of values ​​for each characteristic.

[0028] In some embodiments, the category assigned to a cell of the classification table corresponds to the category of the set of learning data having a maximum number of occurrences of the learning data within a range of values ​​of each characteristic associated with the cell of the classification table.

[0029] In a preferred implementation, if for each set of learning data the number of occurrences of the learning data within the range of values ​​of each characteristic associated with the unit is the same and not zero, the category assigned to the unit of the classification table corresponds to the category with the highest probability compared to the categories assigned to adjacent units of the classification table.

[0030] In a preferred implementation, if for each set of learning data the number of occurrences of the learning data within the range of values ​​of each characteristic associated with the unit is zero, the category assigned to the unit of the classification table corresponds to an undetermined category or to a category with the highest probability compared to the categories assigned to adjacent units of the classification table.

[0031] In some embodiments, the range of values ​​for each property is set using Sturges' rule.

[0032] Advantageously, the method further comprises developing a classification computer program product comprising instructions which, when executed by a computer, cause the computer to implement the generated classification model.

[0033] According to another embodiment, a computer program product is provided. The computer program product includes instructions. When the program is executed by a computer, the instructions cause the computer to implement the method for generating a classification model as described above.

[0034] According to yet another embodiment, there is provided a computer system comprising:

[0035] - a memory comprising a computer program product, the computer program product comprising instructions, which, when executed by a computer, cause the computer to implement the method for generating a classification model as described above,

[0036] - a processing unit configured to execute the computer program product.

[0037] According to yet another embodiment, there is provided a computer-implemented classification method based on a classification model obtained by a generation method as previously described, the classification method comprising:

[0038] - obtain the data to be classified,

[0039] - for each feature identified in the classification model, extract a value from the data to be classified,

[0040] - Determining the category of the data to be classified from a classification table of the classification model, the classification table indicating the categories assigned to the extracted values.

[0041] In an advantageous implementation, the classification model comprises minimum and maximum values ​​for each characteristic studied. The categories assigned to the data to be classified are then found in the classification table according to the values ​​extracted from the data to be classified, the minimum and maximum values ​​for each characteristic studied and the dimensions of each dimension of the classification table.

[0042] According to yet another embodiment, a computer program product is provided. The computer program product includes instructions. When the program is executed by a computer, the instructions enable the computer to implement the classification method as described above.

[0043] According to yet another embodiment, there is provided a computer system comprising:

[0044] - a memory comprising a computer program product, the computer program product comprising instructions, which, when executed by a computer, cause the computer to implement the classification method as described above,

[0045] - a processing unit configured to execute the computer program product. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Other advantages and features of the present invention will become apparent after examining the non-limiting detailed description of embodiments and implementations and the detailed description of the accompanying drawings, in which:

[0047] Figure 1 A computer system configured to implement a method for generating a classification model according to one embodiment is illustrated;

[0048] Figure 2 The embodiment illustrates an implementation of a classification model generation method according to one embodiment;

[0049] Figure 3 illustrates graphs and histograms that may be obtained from values ​​of characteristics of learning data according to one embodiment;

[0050] Figure 4 A computer system configured to implement a classification method according to one embodiment is illustrated; and

[0051] Figure 5 A classification method that can be performed by a computer system according to one embodiment is illustrated. DETAILED DESCRIPTION

[0052] Figure 1An embodiment of a computer system SYS1 configured to implement a method for generating a classification model is illustrated. The computer system SYS1 comprises a processing unit UT1 (e.g., a processor such as a microprocessor, a microcontroller, a central processing unit, etc.) and a memory MEM1. For example, such a computer system SYS1 may be a personal computer or a server.

[0053] The memory MEM1 comprises a computer program PRG1 comprising instructions which, when the program PRG1 is executed by the processing unit UT1 of the computer system SYS1 , cause the processing unit UT1 to implement a method for generating a classification model, as described later.

[0054] The memory MEM1 is also configured to store learning data TDAT. The learning data TDAT is data for generating a classification model MDL. The learning data TDAT is supplied by a user of the computer system SYS1 who wishes to create a classification model based on these learning data TDAT. The learning data TDAT can be obtained from data acquired by a sensor and then supplied to the computer system SYS1. For example, the data TDAT can correspond to any type of analog or digital signal, such as an image, a vibration signal, a current measurement value, etc.

[0055] The learning data TDAT includes data associated with different categories. A category is a set of data that has similar characteristics and is grouped together according to these characteristics. Therefore, the learning data is divided into groups of learning data, each group of learning data is associated with a category indicated by the user.

[0056] More specifically, a feature is a quantitative measurement or property of learning data that allows the learning data to be distinguished from each other and associated with a specific class. The learning data may directly correspond to a feature.

[0057] For example, the characteristics of a time series signal may include amplitude, frequency, duration, minimum value, maximum value, mean value, standard deviation, etc.

[0058] For example, one set of learning data may be associated with a class corresponding to normal data, and another set of learning data may be associated with a class corresponding to abnormal data.

[0059] The processing unit UT1 is configured to generate the classification model MDL by implementing a generation method. In particular, the processing unit UT1 is configured to generate the classification model MDL by executing a computer program PRG1 with the classification data TDAT as input.

[0060] Figure 2 The diagram illustrates an implementation of a classification model generation method, which may be executed by the computer system as described above.

[0061] The method comprises obtaining 20 learning data TDAT associated with different categories. Thus, each category is created using a set of learning data.

[0062] The method comprises identifying 21 at least one characteristic of the learning data to be investigated for classifying the learning data. In particular, one characteristic to be investigated or a plurality of characteristics to be investigated may be identified. These characteristics are set by a user.

[0063] Afterwards, the method comprises extracting 22 the values ​​of the characteristics for the different learning data. If the learning data directly correspond to the characteristics, the extraction step 22 is equivalent to merely reading the learning data.

[0064] Afterwards, the method comprises identifying 23 a range of values ​​for each characteristic of the learning data. Next, these ranges of values ​​will be assigned to different categories.

[0065] In particular, the range of values ​​can be set by segmenting the values ​​extracted from the learning data. For example, the range of values ​​can be set using the Sturges rule or the Freedman-Diaconis rule or the Scott rule known to those skilled in the art.

[0066] A classification model is created based on the number of occurrences of learning data with respect to different set ranges of values.

[0067] More specifically, for each type of learning data, an occurrence table is used to report the number of occurrences of that type of learning data relative to the range of values ​​set for each studied characteristic.

[0068] The occurrence table has a number of dimensions equal to the number of characteristics to be studied. Each dimension of the table has a size corresponding to the number of ranges of values ​​set for the characteristic associated with the dimension. Each cell of the table is associated with a range of values ​​for each characteristic studied. Each cell allows counting the number of occurrences of the learning data of the category associated with the occurrence table relative to the range of values ​​of the characteristic associated with the cell.

[0069] The number of occurrences of learning data relative to a range of values ​​is reported in an occurrence table according to the following method. For each piece of learning data, the value of each studied characteristic extracted from the learning data is compared with the range of values ​​set for the characteristic to determine which cell of the occurrence table should be incremented.

[0070] Each occurrence table is stored in a memory of the computer system, such as a volatile memory of the computer system.

[0071] When studying a single characteristic, each occurrence table can be illustrated by a histogram comprising bins associated with different set ranges of values. Each bin then illustrates the number of occurrences of the learning data of the class associated with the histogram for the range of values ​​of the characteristic under study associated with the bin. Figure 3 Describe such a histogram.

[0072] The classification model MDL to be generated comprises a classification table TAB. In particular, the method comprises creating 24 such a classification table TAB. The classification table TAB has dimensions corresponding to set characteristics. Each cell of the classification table TAB is associated with an arrangement of values ​​for different set characteristics.

[0073] The method comprises assigning 25 a class to each cell of the classification table. In particular, a class is assigned to each cell of the classification table by comparing occurrence tables established for learning data of different classes.

[0074] The category assigned to a cell corresponds to the category of learning data for which the maximum number of occurrences of the learning data is present within the range of the value set for the cell.

[0075] If the number of occurrences of learning data within the range of values ​​of each characteristic associated with a unit is equal and non-zero between different groups of learning data, a unit of the classification table TAB is associated with the class having the highest probability of occurrence compared to all learning data. In particular, the class assigned to such a unit of the classification table can be set based on the class assigned to the adjacent unit of the classification table.

[0076] If, for each set of learning data, the number of occurrences of the learning data within the range of values ​​of each characteristic associated with the unit is zero, the category assigned to the unit of the classification table corresponds to an undetermined category or to a category with the highest probability compared to the categories assigned to adjacent units of the classification table.

[0077] Afterwards, the method comprises a step 26 of generating a classification model. The classification model comprises a classification table and a minimum value Min for each characteristic. val and the maximum value Max val In addition to minimum and maximum values ​​for each feature, a classification model can include a range of values ​​for each feature.

[0078] Such a generation method allows obtaining a classification model that is simple and fast to use and occupies relatively reduced space in the memory.

[0079] Thus, such a classification model can be used by a computer system which comprises limited resources in terms of memory and computing power. In particular, such a classification model can be used by a microcontroller. In particular, the classification model can be used by a computer system such as that described later with reference to Figure 4 The computer system described herein is implemented by a computer system such as the computer system described herein.

[0080] Since categories are assigned to the range of values ​​of each studied characteristic, the classification model is set to improve its generalization to data that has not been used as learning data while avoiding overfitting the learning data.

[0081] Such classification models can also be easily updated by taking into account new learning data.

[0082] Furthermore, if the learning data is divided into several groups of different sizes, weights can be applied to the number of occurrences of the learning data used when assigning categories to different cells of the classification table. Therefore, such a classification model generation method is applicable to unbalanced sets of learning data.

[0083] Figure 3 The graph GRPH and the histograms H_PFFT and H_LINT which can be obtained from the values ​​of the characteristics of the learning data are illustrated.

[0084] In this case, the learning data correspond to signals corresponding to sounds. The characteristics set for these learning data are the peak value of the Fourier transform (particularly the Fast Fourier Transform PFFT), and the curve integral LINT. The first set of learning data is associated with the first category CLS1. The second set of learning data is associated with the second category CLS2.

[0085] Specifically, the abscissa of the graph GRPH corresponds to the peak value of the Fourier transform (specifically the Fast Fourier transform) of the learning data. The ordinate of the graph corresponds to the value of the curve integral of the learning data.

[0086] For each learning data segment associated with category CLS1 or category CLS2, a point is plotted on the graph according to the value of the Fourier transform peak and the curve integral associated with that learning data segment.

[0087] The point P_CLS1 corresponds to the value of the characteristic of the learning data associated with the first class CLS1.

[0088] The point P_CLS2 corresponds to the value of the characteristic of the learning data associated with the second class CLS2.

[0089] The histogram H_PFFT illustrates the number of occurrences of different ranges of the learning data of each class CLS1 , CLS2 according to the value of their Fourier transform peaks relative to the values ​​P_PFFT set for the bins of the histogram.

[0090] The histogram H_LINT illustrates the number of occurrences of different ranges of values ​​of the curve integral of the learning data of each class CLS1 , CLS2 relative to the values ​​P_LINT set for the bins of the histogram.

[0091] If only the Fourier transform peak PFFT is used to generate the classification model, the classification table can be set from the histogram H_PFFT according to the number of occurrences of each bin of the histogram based on the learning data of each category CLS1, CLS2.

[0092] If only the value of the curve integral LINT is used to generate a classification model, a classification table can be set from the histogram H_LINT according to the number of occurrences of each bin of the histogram based on the learning data of each class CLS1, CLS2.

[0093] In order to improve the performance of the classification model, the classification model can be generated by considering the values ​​of the Fourier transform peak PFFT and the curve integral LINT. In this case, the classification table can be set according to the number of occurrences of the learning data of each category CLS1, CLS2 in each area ZNE of the graph GRPH with the size of each range of the set values ​​P_LINT and P_PFFT. Then, each area ZNE is associated with a cell of the previously set occurrence table.

[0094] Figure 4 An embodiment of a computer system SYS2 configured to implement a classification method using the described classification model is illustrated.

[0095] For example, the computer system SYS2 may be a microcontroller, a personal computer or a server.

[0096] The computer system SYS2 comprises a processing unit UT2 and a memory MEM2.

[0097] The computer system SYS2 comprises a computer program PRG2 comprising instructions which, when the program is executed by the processing unit UT2 of the computer system SYS2, cause the processing unit UT2 to implement a classification method described later. The computer program is stored in a memory MEM2.

[0098] The computer system SYS2 is configured to receive data XDAT to be classified.

[0099] Figure 5 The diagram illustrates an implementation of a classification method that can be performed by the computer system SYS2 as described above.

[0100] The method comprises obtaining 30 data XDAT to be classified.

[0101] Thereafter, the method comprises extracting 31 the value of a characteristic from the data to be classified XDAT. The characteristics of the data to be classified correspond to the characteristics set to build the classification model MDL.

[0102] Thereafter, the method comprises determining 32 a category of the data to be classified XDAT using a classification model according to the values ​​of the characteristics extracted from the data to be classified XDAT.

[0103] In particular, the category of the data XDAT to be classified is determined according to the classification table TAB of the classification model MDL. In particular, as previously described, each cell of the classification table TAB is associated with a category.

[0104] The value of the characteristic extracted from the data to be classified XDAT is used to select a cell of the classification table associated with the range of values ​​of the characteristic including the extracted value. This selected cell of the classification table TAB allows determining the category of the data to be classified.

[0105] In particular, the extracted value of the feature and the minimum value of the feature determined by the classification model can be used val and the maximum value Max val To select the cell of the category table TAB.

[0106] In particular, for each characteristic studied, the coordinates of the cells of the classification table TAB can be obtained using the following formula:

[0107] (Dat val -Min val ) / (Max val -Min val / N), where Dat val corresponds to the value of the feature extracted from the data to be classified, Min val corresponds to the minimum value of the feature determined by the classification model, Max val corresponds to the maximum value of the characteristic determined by the classification model MDL, and N corresponds to the number of cells of the classification table associated with the characteristic (ie, the number of value ranges set for the characteristic).

[0108] Alternatively, if the classification model includes all value ranges set for each characteristic during its generation, the cells of the classification table TAB may be selected by comparing the values ​​of the characteristics extracted within the set ranges of values.

Claims

1. A method for generating a classification model, the method comprising: obtaining at least one set of learning data, each set of learning data being associated with an indicated category; identifying at least one characteristic of the learning data to be studied; extracting the value of each feature defined for all learning data; identifying a range of values ​​for each characteristic from among the extracted values; creating a classification table having a number of dimensions corresponding to the number of investigated characteristics, each dimension having a size equal to the number of ranges of values ​​defined for the characteristic associated with that dimension, each cell of the classification table being associated with a range of values ​​for each investigated characteristic; assigning a category to each cell of said classification table according to the number of occurrences of the extracted value of each characteristic of each set of said learning data relative to said range of values ​​defined for each characteristic studied; as well as The classification model is generated, wherein the classification model includes the classification table.

2. The method according to claim 1, wherein the classification model further comprises a minimum value and a maximum value for each characteristic studied.

3. A method according to claim 1, wherein the category assigned to the unit of the classification table corresponds to the category of the group of learning data having the maximum number of occurrences of learning data within the range of the value of each characteristic associated with the unit of the classification table.

4. A method according to claim 1, wherein if, for each set of learning data, the number of occurrences of the learning data within the range of values ​​of each characteristic associated with the unit is the same and not zero, then the category assigned to the unit of the classification table corresponds to the category with the highest probability compared to the category assigned to the adjacent unit of the classification table.

5. A method according to claim 1, wherein if, for each set of learning data, the number of occurrences of the learning data within the range of values ​​of each characteristic associated with the unit is zero, then the category assigned to the unit of the classification table corresponds to an undetermined category or to the category having the highest probability compared to the category assigned to the adjacent unit of the classification table. The method of claim 1 , wherein the range of the values ​​for each characteristic is set using Sturges' rule.

7. The method according to claim 1, further comprising: A classification computer program product is developed, the classification computer program product comprising instructions which, when executed by a computer, cause the computer to implement the generated classification model.

8. A computer system comprising: A memory comprising a computer program, the computer program comprising instructions to: obtaining at least one set of learning data, each set of learning data being associated with an indicated category; identifying at least one characteristic of the learning data to be studied; extracting the value of each feature defined for all learning data; identifying a range of values ​​for each characteristic from among the extracted values; creating a classification table having a number of dimensions corresponding to the number of investigated characteristics, each dimension having a size equal to the number of ranges of values ​​defined for the characteristic associated with that dimension, each cell of the classification table being associated with a range of values ​​for each investigated characteristic; assigning a category to each cell of said classification table according to the number of occurrences of the extracted value of each characteristic of each set of said learning data relative to said range of values ​​defined for each characteristic studied; as well as generating a classification model, wherein the classification model includes the classification table; as well as A processor is configured to execute the computer program.

9. The computer system of claim 8, wherein the classification model further comprises a minimum value and a maximum value for each characteristic studied.

10. A computer system according to claim 8, wherein the category assigned to the cell of the classification table corresponds to the category of the group of learning data having the maximum number of occurrences of learning data within the range of the value of each characteristic associated with the cell of the classification table.

11. A computer system according to claim 8, wherein if, for each set of learning data, the number of occurrences of the learning data within the range of values ​​of each characteristic associated with the unit is the same and not zero, then the category assigned to the unit of the classification table corresponds to the category with the highest probability compared to the category assigned to the adjacent unit of the classification table.

12. A computer system according to claim 8, wherein if, for each set of learning data, the number of occurrences of the learning data within the range of values ​​of each characteristic associated with the unit is zero, then the category assigned to the unit of the classification table corresponds to an undetermined category or to the category having the highest probability compared to the category assigned to the adjacent unit of the classification table.

13. The computer system of claim 8, wherein the range of the values ​​for each characteristic is set using Sturges' rule.

14. The computer system of claim 8, further comprising developing a classification computer program product, the classification computer program product comprising instructions that, when the program is executed by a computer, cause the computer to implement the generated classification model.

15. A classification method, comprising: obtaining at least one set of learning data, each set of learning data being associated with an indicated category; identifying at least one characteristic of the learning data to be studied; extracting the value of each feature defined for all learning data; identifying a range of values ​​for each characteristic from among the extracted values; creating a classification table having a number of dimensions corresponding to the number of investigated characteristics, each dimension having a size equal to the number of ranges of values ​​defined for the characteristic associated with that dimension, each cell of the classification table being associated with a range of values ​​for each investigated characteristic; assigning a category to each cell of said classification table according to the number of occurrences of the extracted value of each characteristic of each set of said learning data relative to said range of values ​​defined for each characteristic studied; generating a classification model, wherein the classification model includes the classification table; Obtaining data to be classified; extracting a value from the data to be classified for each characteristic defined in the classification model; as well as The category of the data to be classified is determined from the classification table of the classification model, the classification table indicating the category assigned to the extracted value.

16. A method according to claim 15, wherein the classification model includes minimum and maximum values ​​for each studied characteristic, and wherein the categories assigned to the data to be classified are found in the classification table based on the values ​​extracted from the data to be classified, the minimum and maximum values ​​for each studied characteristic and the size of each dimension of the classification table.

17. The method of claim 15, wherein the classification model further comprises a minimum value and a maximum value for each characteristic studied.

18. A method according to claim 15, wherein the category assigned to the unit of the classification table corresponds to the category of the group of learning data having the maximum number of occurrences of learning data within the range of the value of each characteristic associated with the unit of the classification table.

19. A method according to claim 15, wherein if, for each set of learning data, the number of occurrences of the learning data within the range of values ​​of each characteristic associated with the unit is the same and not zero, then the category assigned to the unit of the classification table corresponds to the category with the highest probability compared to the category assigned to the adjacent unit of the classification table.

20. A method according to claim 15, wherein if, for each set of learning data, the number of occurrences of the learning data within the range of values ​​of each characteristic associated with the unit is zero, then the category assigned to the unit of the classification table corresponds to an undetermined category or to the category having the highest probability compared to the category assigned to the adjacent unit of the classification table.

Images