Inspector behavior risk identification method based on data mining

Abstract

The present invention relates to the field of data processing technology, and in particular to a method for identifying the behavior risks of inspectors based on data mining, comprising: step S1, collecting inspector behavior data; step S2, organizing the inspector behavior data into a csv file; step S3, performing aggregation analysis on the csv file; step S4, performing normalization processing; step S5, encoding the time series signal of the normalized feature vector through a recursive network neural model; step S6, dividing the encoded time series signal into a training set and a test set; step S7, training an SVM classifier; step S8, testing the trained SVM classifier; step S9, judging whether the training of the SVM classifier meets the standard; step S10, correcting the parameters of the recursive network model; step S11, inputting the real-time collected inspector behavior data into an inspector behavior risk identification model. The present invention improves the efficiency and accuracy of used car inspection.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method for identifying behavioral risks of inspectors based on data mining. Background Art

[0002] For sellers, inspections can provide detailed information about the vehicle and help them understand its condition, allowing them to set a reasonable selling price. For buyers, inspections can reveal potential problems with the vehicle and help them avoid purchasing vehicles with serious hidden dangers. Artificial intelligence technology provides powerful algorithms and computing power for behavioral feature extraction. Through deep learning, machine learning and other technologies, we can train models to automatically learn and extract behavioral features from large amounts of data. These models are able to identify and analyze complex behavioral patterns, thereby more accurately describing and understanding human behavior, improving the accuracy and reliability of used car inspections and further enhancing consumers' trust in used car inspection services.

[0003] Chinese Patent Publication No. CN108520237B discloses a risk identification method, which includes: obtaining the body shape of a target person in any video frame in a video stream sequence; wherein the body shape of the target person is positional information of each body part of the target person; obtaining motion information of the target person in the video frame based on the body shape of the target person in the video frame and the body shape of the target person in the previous video frame; wherein the motion information of the target person is motion information of each body part of the target person; and obtaining a risk behavior identification result of the target person based on the body shape and motion information of the target person in a first preset number of video frames. However, this solution cannot be applied to used car inspection platforms and fails to improve the efficiency and accuracy of used car inspections. Summary of the Invention

[0004] To this end, the present invention provides a method for identifying inspector behavior risks based on data mining to overcome the problems of low efficiency and accuracy in used car inspection in the prior art.

[0005] To achieve the above objectives, the present invention provides a method for identifying behavioral risks of inspectors based on data mining, comprising:

[0006] Step S1, collecting the tester's test behavior data;

[0007] Step S2, organizing the inspector behavior data into a csv file;

[0008] Step S3, performing aggregation analysis on the CSV file to obtain a feature vector;

[0009] Step S4, normalizing the feature vector;

[0010] Step S5, constructing a recursive neural network model, and encoding the time series signal of the normalized feature vector through the recursive neural network model;

[0011] Step S6, dividing the encoded time series signal into a training set and a test set;

[0012] Step S7, training the SVM classifier using the training set, and outputting the trained SVM classifier;

[0013] Step S8, testing the trained SVM classifier using the test set and outputting the test accuracy;

[0014] Step S9: Compare the test accuracy with the preset accuracy, determine the training compliance of the SVM classifier based on the comparison result, and output the SVM classifier that has been trained to the compliance as the inspector behavior risk identification model;

[0015] Step S10, correcting the parameters of the recursive network model according to the parameter correction coefficient to obtain corrected training data, and repeating steps S6 to S9 based on the trained data;

[0016] Step S11: input the inspector behavior data collected in real time into the inspector behavior risk identification model, and output the inspector behavior risk identification result.

[0017] Furthermore, in step S2, the inspector behavior data is organized into a CSV file containing the work order code, accident type, action code and action occurrence time through a data processing tool.

[0018] Furthermore, in step S3, the CSV file is imported into a pandas DataFrame, and the CSV file is aggregated and analyzed using a groupby method and an aggregation function to obtain a feature vector.

[0019] Furthermore, in step S4, the feature vector is normalized by a global maximum and minimum normalization method, a category maximum and minimum normalization method, and a size sorting index value normalization method, wherein:

[0020] The formula of the global maximum and minimum normalization method is Xnor = (X-Xmin) / (Xmax-Xmin), where X is the value of the eigenvector, Xnor is the value of the normalized eigenvector, Xmin is the minimum value of x in the global data set, and Xmax is the maximum value of x in the global data set.

[0021] The formula for the category maximum and minimum normalization method is Xnor = (X-Xmin_c l ass) / (Xmax_c l ass-Xmin_c l ass), where Xmin_c l ass is the minimum value of X in the category to which it belongs, and Xmax_c l ass is the maximum value of X in the category to which it belongs;

[0022] The formula for the normalization method of the size sorting index value is Xnor = Xindex / list_size, where Xindex is the sequential index value after sorting the features by size, and list_size is the total number of features involved in the sorting.

[0023] Furthermore, in step S5, when constructing the recursive neural network model, the training cycle is set to E0=10, the batch size is set to B0=32, the output layer is set to 63, and the timing signal of the preset feature vector is divided into a 70% timing signal training set and a 30% timing signal test set. The timing signal training set is input into the recursive neural network model for training, and the timing signal test set is input into the trained recursive neural network model. The parameters in the recursive neural network model are optimized and iterated until the accuracy of the output result of the timing signal test set of the recursive neural network model reaches 98%. The recursive neural network model is output, and the timing signal of the normalized feature vector is input into the recursive neural network model. The recursive neural network model encodes the timing signal of the normalized feature vector to obtain the encoded timing signal.

[0024] Furthermore, in step S6, the encoded time series signal is divided into a 70% training set and a 30% test set.

[0025] Furthermore, in step S7, the parameters of the SVM classifier are set to a linear kernel, the training set is input into the SVM classifier, the SVM classifier is trained, and the trained SVM classifier is output.

[0026] Furthermore, in step S8, the test set is input into the trained SVM classifier for testing, and the test accuracy H is output.

[0027] Furthermore, in step S9, the test accuracy K is compared with the preset accuracy K0, and the training compliance of the SVM classifier is judged based on the comparison result, wherein:

[0028] When K < K0, it is determined that the training of the SVM classifier does not meet the standards;

[0029] When K≥K0, the training of the SVM classifier is determined to have met the standards, and the SVM classifier is output as the inspector behavior risk identification model.

[0030] Furthermore, in step S10, when the training of the SVM classifier does not meet the standard, the parameters of the recursive network model are corrected according to the parameter correction coefficient to obtain the corrected training data, and steps S6 to S9 are repeated according to the trained data, wherein the parameter correction coefficient P is set to P = 0.4 + 0.8 × e -0.3×(E0-B0) , e is the base of the natural logarithm, and the training period E0 and batch size B0 are corrected according to the parameter correction coefficient. The corrected training period is set to Ep0, Ep0 = EO×P, and the corrected batch size is set to Bp0, Bp0 = B0×P.

[0031] Compared with the prior art, the beneficial effect of the present invention is that the method collects the inspector's inspection behavior data through step S1 to improve the accuracy of the data, the method organizes the inspector's behavior data into a csv file through step S2, and performs aggregation analysis on the csv file through step S3 to obtain a feature vector, so as to facilitate subsequent processing and analysis of the data, thereby improving the efficiency of used car inspection, the method normalizes the feature vector through step S4, and also constructs a recursive neural network model through step S5, and encodes the time series signal of the normalized feature vector through the recursive network neural model, thereby improving the robustness of the data, thereby improving the accuracy of used car inspection, the method divides the encoded time series signal into a training set and a test set through step S6, so as to facilitate subsequent training and testing of the SVM classifier, the method trains the SVM classifier through the training set through step S7, outputs the trained SVM classifier, and passes through step S8 The test set tests the trained SVM classifier and outputs the test accuracy to facilitate subsequent evaluation of the detection effect, thereby improving the accuracy of used car detection. The method compares the test accuracy with the preset accuracy through step S9, judges the training compliance of the SVM classifier based on the comparison result, and outputs the trained SVM classifier as an inspector behavior risk identification model to improve the efficiency and accuracy of used car detection. The method corrects the parameters of the recursive network model according to the parameter correction coefficient through step S10 to obtain the corrected training data, and repeats steps S6 to S9 based on the trained data to improve the accuracy of the data, thereby further improving the accuracy of used car detection. The method inputs the real-time collected inspector behavior data into the inspector behavior risk identification model through step S11, and outputs the inspector behavior risk identification result, so as to improve the data accuracy and timeliness, thereby improving the efficiency and accuracy of used car detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 This is a flow chart of the method for identifying behavioral risks of inspectors based on data mining in this embodiment. DETAILED DESCRIPTION

[0033] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are merely used to explain the present invention and are not intended to limit the present invention.

[0034] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0035] See also Figure 1 As shown, it is a flow chart of the method for identifying the behavioral risks of inspectors based on data mining in this example, and the method includes:

[0036] Step S1, collecting the tester's test behavior data;

[0037] Step S2, organizing the inspector behavior data into a csv file;

[0038] Step S3, performing aggregation analysis on the CSV file to obtain a feature vector;

[0039] Step S4, normalizing the feature vector;

[0040] Step S5, constructing a recursive neural network model, and encoding the time series signal of the normalized feature vector through the recursive neural network model;

[0041] Step S6, dividing the encoded time series signal into a training set and a test set;

[0042] Step S7, training the SVM classifier using the training set, and outputting the trained SVM classifier;

[0043] Step S8, testing the trained SVM classifier using the test set and outputting the test accuracy;

[0044] Step S9: Compare the test accuracy with the preset accuracy, determine the training compliance of the SVM classifier based on the comparison result, and output the SVM classifier that has been trained to the compliance as the inspector behavior risk identification model;

[0045] Step S10, correcting the parameters of the recursive network model according to the parameter correction coefficient to obtain corrected training data, and repeating steps S6 to S9 based on the trained data;

[0046] Step S11: input the inspector behavior data collected in real time into the inspector behavior risk identification model, and output the inspector behavior risk identification result.

[0047] Specifically, the method is applied to a used car inspection platform. The time series signal of the normalized feature vector is encoded through a recursive neural network model to facilitate subsequent data analysis and processing. The trained SVM classifier is output as an inspector behavior risk identification model to improve data processing efficiency, thereby improving the efficiency of used car inspection.

[0048] Specifically, the method collects the inspector's inspection behavior data through step S1 to improve the accuracy of the data, organizes the inspector's behavior data into a csv file through step S2, and performs aggregation analysis on the csv file through step S3 to obtain a feature vector, so as to facilitate subsequent data processing and analysis, thereby improving the efficiency of used car inspection, the method normalizes the feature vector through step S4, and also constructs a recursive neural network model through step S5, and encodes the time series signal of the normalized feature vector through the recursive network neural model, thereby improving the robustness of the data, thereby improving the accuracy of used car inspection, the method divides the encoded time series signal into a training set and a test set through step S6, so as to facilitate subsequent training and testing of the SVM classifier, the method trains the SVM classifier through the training set through step S7, outputs the trained SVM classifier, and tests the trained SVM classifier through the test set through step S8. The SVM classifier is tested and the test accuracy is output to facilitate subsequent evaluation of the detection effect, thereby improving the accuracy of used car detection. The method compares the test accuracy with the preset accuracy through step S9, judges the training compliance of the SVM classifier based on the comparison result, and outputs the trained SVM classifier as an inspector behavior risk identification model to improve the efficiency and accuracy of used car detection. The method corrects the parameters of the recursive network model according to the parameter correction coefficient through step S10 to obtain the corrected training data, and repeats steps S6 to S9 based on the trained data to improve the accuracy of the data, thereby further improving the accuracy of used car detection. The method inputs the real-time collected inspector behavior data into the inspector behavior risk identification model through step S11, and outputs the inspector behavior risk identification result, so as to improve the data accuracy and timeliness, thereby improving the efficiency and accuracy of used car detection.

[0049] Specifically, in step S1, the tester's test behavior data is collected by testing the test log and embedded data in the test APP;

[0050] Specifically, the inspection APP refers to a mobile application that provides inspectors with functions such as used car status, vehicle information, and vehicle value assessment. This embodiment does not limit the type of inspection APP. Relevant technical personnel in this field can freely set it according to actual needs. It only needs to meet the needs of collecting inspector inspection behavior data. For example, the inspection APP can be set to Dr. Cha. The inspection log refers to the historical records of inspection activities, inspection results and other relevant information recorded by the system and software during the inspection process. The buried data refers to the process of collecting and recording data for all click actions of the inspector in the inspection APP, such as clicking buttons, links, pictures and menu items.

[0051] Specifically, in step S2, the inspector behavior data is organized into a CSV file containing the work order code, accident type, action code and action occurrence time through a data processing tool.

[0052] Specifically, the data processing tool refers to software and hardware tools used to collect, store, process, analyze and display data. This embodiment does not limit the type of data processing tool. Relevant technical personnel in this field can freely set it according to actual needs. It only needs to meet the requirements of organizing the inspector's behavior data into a CSV file containing work order code, accident type, action code and action occurrence time. For example, the data processing tool can be set as a database management system DBMS. The work order code refers to a unique identifier used to uniquely identify and track work orders or work orders. The accident type refers to the classification of problems and anomalies found by the inspector during the inspection process, such as dividing the accident type into "1: accident vehicle" and "2: non-accident vehicle". The action code refers to the code used to identify the specific actions and steps performed by the inspector during the inspection process. The action occurrence time refers to the specific time when the inspector performs a specific action or step. The CSV file refers to a spreadsheet file format, which is used to store data in tabular form.

[0053] Specifically, in step S3, the CSV file is imported into a pandas DataFrame, and the CSV file is aggregated and analyzed using a groupby method and an aggregation function to obtain a feature vector.

[0054] Specifically, the pandas DataFrame refers to a core data structure in the pandas library for storing and manipulating two-dimensional, variable-size tabular data that can contain columns of heterogeneous types. The groupby method refers to a method for grouping data according to one or more specified keys and returning a special object. The special object refers to the object returned after using the groupby method. The aggregate function refers to a function used to perform calculations on a set of values ​​and return a single value. This embodiment does not limit the type of aggregate function. Relevant technical personnel in this field can freely set it according to actual needs, and it only needs to meet the needs of aggregate analysis of CSV files. For example, the type of aggregate function can be set to a sum function. The eigenvector refers to a non-zero vector that is only scaled under linear transformation.

[0055] Specifically, in step S4, the feature vector is normalized by a global maximum and minimum normalization method, a category maximum and minimum normalization method, and a size sorting index value normalization method, wherein:

[0056] The formula of the global maximum and minimum normalization method is Xnor = (X-Xmin) / (Xmax-Xmin), where X is the value of the eigenvector, Xnor is the value of the normalized eigenvector, Xmin is the minimum value of x in the global data set, and Xmax is the maximum value of x in the global data set.

[0057] The formula for the category maximum and minimum normalization method is Xnor = (X-Xmin_c l ass) / (Xmax_c l ass-Xmin_c l ass), where Xmin_c l ass is the minimum value of X in the category to which it belongs, and Xmax_c l ass is the maximum value of X in the category to which it belongs;

[0058] The formula for the normalization method of the size sorting index value is Xnor = Xindex / list_size, where Xindex is the sequential index value after sorting the features by size, and list_size is the total number of features involved in the sorting.

[0059] Specifically, the global maximum and minimum normalization method refers to a normalization method that linearly maps data to the interval [0, 1]. The category maximum and minimum normalization method refers to a method that performs maximum and minimum normalization on each category of the data set. The size sort index value normalization method refers to a normalization method based on the data sort position. The normalization process refers to a data preprocessing method that converts the actual variable value to an appropriate range. The value of the eigenvector refers to the eigenvalue of each sample or observation in the original data set. The value of the normalized eigenvector refers to the eigenvalue obtained after normalization. The minimum value of the global data set refers to the minimum value of all features in the entire data set. The maximum value of the global data set refers to the maximum value of all features in the entire data set. The minimum value of the category refers to the minimum value of all features in the original data set. The maximum value of the category refers to the maximum value of all features in the original data set. The sequential index value after sorting the features by size refers to the position index of each eigenvalue in the sorted list after sorting the eigenvalues. The total number of features involved in the sorting refers to the total number of eigenvalues ​​considered when performing the sorting operation.

[0060] Specifically, in step S5, when constructing the recursive neural network model, the training cycle is set to E0=10, the batch size is set to B0=32, the output layer is set to 63, and the time series signal of the preset feature vector is divided into a 70% time series signal training set and a 30% time series signal test set. The time series signal training set is input into the recursive neural network model for training, and the time series signal test set is input into the trained recursive neural network model. The parameters in the recursive neural network model are optimized and iterated until the accuracy of the output result of the time series signal test set of the recursive neural network model reaches 98%. The recursive neural network model is output, and the time series signal of the normalized feature vector is input into the recursive neural network model. The recursive neural network model encodes the time series signal of the normalized feature vector to obtain the encoded time series signal.

[0061] Specifically, the training cycle refers to the number of times the training set is completely traversed by the forward propagation and backpropagation algorithms, the batch size refers to the number of samples processed by the model before each update of the model parameters, the output layer refers to the last layer of the recursive neural network model, the time series signal of the preset feature vector refers to the feature vector preset in time sequence for training the recursive neural network model, the time series signal refers to a series of data points that change over time, and the encoding refers to the process of converting the time series signal into a more compact and easier to process form.

[0062] Specifically, in step S6, the encoded time series signal is divided into a 70% training set and a 30% test set.

[0063] Specifically, the training set refers to a data set used to train a machine learning model, and the test set refers to a data set used to evaluate the performance of a machine learning model.

[0064] Specifically, in step S7, the parameters of the SVM classifier are set to a linear kernel, the training set is input into the SVM classifier, the SVM classifier is trained, and the trained SVM classifier is output.

[0065] Specifically, the SVM classifier refers to an instance of a support vector machine classification algorithm, and the linear kernel refers to a kernel function in a support vector machine for solving linearly separable classification problems. The mathematical expression of the linear kernel is set to K(x,y)=x·y, where x and y are feature vectors and “·” represents the dot product of two vectors.

[0066] Specifically, in step S8, the test set is input into the trained SVM classifier for testing, and the test accuracy H is output.

[0067] Specifically, the test accuracy H refers to the ratio of the number of correctly predicted samples Q predicted by the SVM classifier on the test set to the total number of test samples Q0 when the trained SVM classifier is tested, that is, H=Q / Q0.

[0068] Specifically, in step S9, the test accuracy K is compared with the preset accuracy K0, and the training compliance of the SVM classifier is judged based on the comparison result, wherein:

[0069] When K < K0, it is determined that the training of the SVM classifier does not meet the standards;

[0070] When K≥K0, the training of the SVM classifier is determined to have met the standards, and the SVM classifier is output as the inspector behavior risk identification model.

[0071] Specifically, the preset accuracy rate K0 refers to a preset value to reflect the training compliance of the SVM classifier. This embodiment does not limit the value of the preset accuracy rate K0. Relevant technical personnel in this field can freely set it according to actual needs. It only needs to meet the requirements of reflecting the training compliance of the SVM classifier. For example, the preset accuracy rate K0 can be set to 0.6.

[0072] Specifically, in step S10, when the training of the SVM classifier does not meet the standard, the parameters of the recursive network model are corrected according to the parameter correction coefficient to obtain the corrected training data, and steps S6 to S9 are repeated according to the trained data, wherein the parameter correction coefficient P is set to P = 0.4 + 0.8 × e -0.3×(E0-B0), e is the base of the natural logarithm, and the training period E0 and batch size B0 are corrected according to the parameter correction coefficient. The corrected training period is set to Ep0, Ep0 = EO×P, and the corrected batch size is set to Bp0, Bp0 = B0×P.

[0073] Specifically, the parameter correction coefficient refers to a coefficient used to correct the training cycle E0 and the batch size B0.

[0074] Specifically, in step S11, the inspector behavior data collected in real time is input into the inspector behavior risk identification model, and the inspector behavior risk identification result is output.

[0075] Specifically, the inspector behavior risk identification result refers to the result obtained by the inspector behavior risk identification model through model calculation based on the inspector behavior data collected in real time. The inspector behavior risk identification result includes accident vehicle inspection behavior and non-accident vehicle inspection behavior.

[0076] Thus far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art may make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will fall within the scope of protection of the present invention.

Claims

1. A method for identifying behavioral risks of inspectors based on data mining, characterized in that: include: Step S1, collecting the tester's test behavior data; Step S2, organizing the inspector behavior data into a csv file; Step S3, performing aggregation analysis on the CSV file to obtain a feature vector; Step S4, normalizing the feature vector; Step S5, constructing a recursive neural network model, and encoding the time series signal of the normalized feature vector through the recursive neural network model; Step S6, dividing the encoded time series signal into a training set and a test set; Step S7, training the SVM classifier using the training set, and outputting the trained SVM classifier; Step S8, testing the trained SVM classifier using the test set and outputting the test accuracy; Step S9: Compare the test accuracy with the preset accuracy, determine the training compliance of the SVM classifier based on the comparison result, and output the SVM classifier that has been trained to the compliance as the inspector behavior risk identification model; Step S10, correcting the parameters of the recursive network model according to the parameter correction coefficient to obtain corrected training data, and repeating steps S6 to S9 based on the trained data; Step S11: input the inspector behavior data collected in real time into the inspector behavior risk identification model, and output the inspector behavior risk identification result.

2. The method for identifying inspector behavior risks based on data mining according to claim 1, characterized in that: In step S2, the inspector behavior data is organized into a CSV file containing the work order code, accident type, action code and action occurrence time through a data processing tool.

3. The method for identifying inspector behavior risks based on data mining according to claim 2, characterized in that: In step S3, the CSV file is imported into a pandas DataFrame, and the CSV file is aggregated and analyzed using a groupby method and an aggregation function to obtain a feature vector.

4. The method for identifying inspector behavior risks based on data mining according to claim 1, characterized in that: In step S4, the feature vector is normalized by using the global maximum and minimum normalization method, the category maximum and minimum normalization method, and the size sorting index value normalization method, wherein: The formula of the global maximum and minimum normalization method is Xnor = (X-Xmin) / (Xmax-Xmin), where X is the value of the eigenvector, Xnor is the value of the normalized eigenvector, Xmin is the minimum value of x in the global data set, and Xmax is the maximum value of x in the global data set. The formula for the category maximum and minimum normalization method is Xnor = (X-Xmin_class) / (Xmax_class-Xmin_class), where Xmin_class is the minimum value of X in the category to which it belongs, and Xmax_class is the maximum value of X in the category to which it belongs; The formula for normalizing the size sort index value is Xnor = Xindex / list_size, where Xindex is the sequential index value after sorting the features by size, and list_size is the total number of features involved in the sorting.

5. The method for identifying inspector behavior risks based on data mining according to claim 1 is characterized in that: In step S5, when constructing the recursive neural network model, the training cycle is set to E0=10, the batch size is set to B0=32, the output layer is set to 63, and the time series signal of the preset feature vector is divided into a 70% time series signal training set and a 30% time series signal test set. The time series signal training set is input into the recursive neural network model for training, and the time series signal test set is input into the trained recursive neural network model. The parameters in the recursive neural network model are optimized and iterated until the accuracy of the output result of the time series signal test set of the recursive neural network model reaches 98%. The recursive neural network model is output, and the time series signal of the normalized feature vector is input into the recursive neural network model. The recursive neural network model encodes the time series signal of the normalized feature vector to obtain the encoded time series signal.

6. The method for identifying inspector behavior risks based on data mining according to claim 1, characterized in that: In step S6, the encoded time series signal is divided into a 70% training set and a 30% test set.

7. The method for identifying inspector behavior risks based on data mining according to claim 1, characterized in that: In step S7, the parameters of the SVM classifier are set to a linear kernel, the training set is input into the SVM classifier, the SVM classifier is trained, and the trained SVM classifier is output.

8. The method for identifying inspector behavior risks based on data mining according to claim 1, characterized in that: In step S8, the test set is input into the trained SVM classifier for testing, and the test accuracy H is output.

9. The method for identifying inspector behavior risks based on data mining according to claim 1, characterized in that: In step S9, the test accuracy K is compared with the preset accuracy K0, and the training compliance of the SVM classifier is judged based on the comparison result, wherein: When K < K0, it is determined that the training of the SVM classifier does not meet the standards; When K≥K0, the training of the SVM classifier is determined to have met the standards, and the SVM classifier is output as the inspector behavior risk identification model.

10. The method for identifying inspector behavior risks based on data mining according to claim 9, characterized in that: In step S10, when the training of the SVM classifier does not meet the standard, the parameters of the recursive network model are corrected according to the parameter correction coefficient to obtain the corrected training data, and steps S6 to S9 are repeated according to the trained data, wherein the parameter correction coefficient P is set to P = 0.4 + 0.8 × e -0.3×(E0-B0) , e is the base of the natural logarithm, and the training period E0 and batch size B0 are corrected according to the parameter correction coefficient. The corrected training period is set to Ep0, Ep0 = EO×P, and the corrected batch size is set to Bp0, Bp0 = B0×P.

Images