A method and system for detecting incomplete track of business process event log

Abstract

The present invention discloses a method and system for detecting incomplete tracks of business process event logs, including: 1) obtaining standardized event logs; 2) extracting tracks existing in the standardized event logs according to the standardized event logs to establish a data set, and dividing the data set into a training set and a test set; 3) extracting features of the training set according to the last event and the frequency of occurrence of the event; 4) performing unique hot encoding on the extracted features and updating them to the training set; 5) inputting the encoded feature values ​​into a CatBoost machine learning model for iterative training and parameter optimization, thereby obtaining an optimal CatBoost machine learning model; 6) inputting the test set into the optimal CatBoost machine learning model for testing, and finally detecting the incomplete tracks existing in the data set. The present invention improves the available quality of the event logs by detecting incomplete tracks in the business process event logs, and lays a foundation for modeling and optimization of subsequent business processes.

Description

Technical Field

[0001] The present invention relates to the technical field of business process management, and in particular to a method and system for detecting incomplete traces of business process event logs. Background Art

[0002] Market competition has forced enterprises to change their traditional management and operation methods and establish information system management mechanisms within the enterprise to support the execution of business processes and the recording of event logs. In recent years, process mining technology has enabled information systems to better analyze their business processes, helping enterprises provide new monitoring and improvement methods to better adapt to and meet the ever-changing market needs.

[0003] Process discovery is the core task of process mining, and its purpose is to discover process models from the event logs generated by the operation of business processes. Current process mining algorithms all assume that the trajectories contained in the event logs are complete. However, in real production life, due to external factors such as system crashes, data acquisition instrument failures, or data transmission, log data may be lost, and the logs may be incomplete or reflect other incorrect behaviors. The existence of incomplete data will reduce the reliability and usability of event logs, thereby having a negative impact on the results of process mining.

[0004] In order to limit its negative impact, event logs are usually preprocessed. However, preprocessing is a challenging and time-consuming task, especially in the case of large logs containing complex behaviors. Most of the related works on event log detection and filtering require users to have prior knowledge of the process mining field in order to return results. Therefore, potential improvements lie in finding a general technique to discover and explain incorrect behaviors in processes. Summary of the invention

[0005] The first purpose of the present invention is to overcome the shortcomings and deficiencies of the prior art and propose a method for detecting incomplete trajectories in business process event logs, which can accurately detect incomplete trajectories in logs, thereby improving the available quality of event logs and laying the foundation for the management and optimization of business processes.

[0006] A second object of the present invention is to provide a business process event log incomplete track detection system.

[0007] The first object of the present invention is achieved by the following technical solution: A method for detecting incomplete traces of business process event logs, comprising the following steps:

[0008] 1) Obtain a standardized event log, which records the relevant business processes, including trajectories, events, and event timestamps;

[0009] 2) According to the standardized event log obtained in step 1), the trajectories existing in the standardized event log are extracted to establish a data set, and the data set is divided into a training set and a test set;

[0010] 3) Extracting features from the training set obtained in step 2) according to the last event and the frequency of occurrence of the event;

[0011] 4) One-hot encode the features extracted in step 3) and update them to the training set so that the encoded feature values ​​can represent the features;

[0012] 5) Inputting the encoded feature values ​​in step 4) into the CatBoost machine learning model for iterative training, and optimizing the model parameters, after reaching the set number of iterations, obtaining the optimal CatBoost machine learning model;

[0013] 6) Input the test set into the optimal CatBoost machine learning model obtained in step 5) for testing, and finally detect the incomplete trajectories in the data set.

[0014] Furthermore, in step 1), redundant attributes in the event log are deleted, and only the trace, event, and event timestamp are retained, thereby forming a standardized event log; all event logs are obtained through the 4TU platform.

[0015] Further, in step 2), according to the standardized event log obtained in step 1), the trajectories in the standardized event log are extracted and a data set is constructed. The specific steps are as follows:

[0016] 2.1) Traverse the event log and extract the trajectory when the trajectory identifier is accessed;

[0017] 2.2) Establishing a set E, inputting the trajectories extracted in step 2.1) into the set E to establish a data set;

[0018] 2.3) The data set obtained in step 2.2) is divided into 70% as the training set and the remaining 30% as the test set.

[0019] Further, in step 3), the training set obtained in step 2) is subjected to feature extraction according to the last event and the frequency of event occurrence, as follows:

[0020] Last event: traverse each track, and when the last event identifier is accessed, extract the event and record it in the end event list;

[0021] Event frequency: traverse each trajectory and count the events in the trajectory. When the same event is accessed, the counter is increased by 1. After each trajectory is traversed, the maximum value of the counter is recorded in the event list.

[0022] Furthermore, in step 4), the one-hot encoding method is used to encode the features extracted in step 3), and the encoded features are updated to the training set processed in step 2), wherein the one-hot encoding maps the categorical variables to integer values ​​and represents each value in the form of a binary vector.

[0023] Furthermore, in step 5), the last event encoded in step 4) and two different types of feature values ​​of event occurrence frequency are input into the CatBoost machine learning model for iterative training, that is, firstly, an input sample and a target value are generated for each prefix length of the trajectory, the input sample represents the current state of the process in the model, and the target value is a binary category label. When the current trajectory is detected to be an incomplete trajectory, the target sample is "true", otherwise it is "false". Once the input sample is mapped to the feature space, it can be fed back to the model together with the target sample; finally, the model quality is evaluated according to the area under the curve obtained by model training, and the model parameters are adjusted to obtain the optimal CatBoost machine learning model with different characteristics.

[0024] Further, in step 6), the test set is input into the optimal CatBoost machine learning model with different features obtained in step 5) to detect incomplete trajectories existing in the data set under different features.

[0025] The second object of the present invention is achieved by the following technical solution: A business process event log incomplete track detection system, comprising:

[0026] The data preprocessing module is used to delete redundant attributes in the event log and retain only the track, event and event timestamp, thereby forming a standardized event log, which records the relevant business processes;

[0027] A data generation module is used to extract and construct a trajectory data set existing in the standardized event log obtained by the data preprocessing module, and divide the data set into a training set and a test set;

[0028] Feature extraction module, used to extract the features of the last event and the frequency of event occurrence in the training set;

[0029] A feature encoding module is used to perform targeted encoding on the features in the feature extraction module;

[0030] The model training module is used to input the feature values ​​obtained by the feature encoding module into the CatBoost machine learning model for iterative training. After reaching the set number of iterations, the optimal CatBoost machine learning model is obtained;

[0031] The detection result output module is used to input the test set into the optimal CatBoost machine learning model obtained by the model training module for verification, so as to detect whether there are incomplete trajectories in the data set and output the detection results.

[0032] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0033] 1. The present invention uses a highly accurate machine learning algorithm for detection, which significantly improves detection accuracy compared to traditional methods, thereby providing business process managers with a more favorable decision-making basis.

[0034] 2. The present invention trains a model through machine learning methods to detect incomplete behaviors in event logs. By extracting feature information that has a decisive influence on the detection task, incorrect behaviors in the process can be discovered. No additional data structure is required, thereby improving program efficiency.

[0035] 3. Without the need for prior knowledge of the process mining field, the present invention proposes a universal incomplete trajectory detection and classification technology, which can reduce the complexity of event logs and provide support for further mining and transformation of the model.

[0036] 4. The present invention can help business process managers extract useful knowledge without increasing computing costs, thereby improving the quality of existing log data modeling.

[0037] 5. The present invention has a wide range of applications in business process management and optimization tasks, is simple to operate, and has strong adaptability, and has broad prospects in improving the quality of event log availability. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 The figure is a logical flow chart of the method of the present invention.

[0039] Figure 2 Schematic diagram of feature extraction of the present invention.

[0040] Figure 3 This is a framework diagram of incomplete trajectory detection based on machine learning algorithm in the present invention.

[0041] Figure 4 This is a comparison diagram of the incomplete trajectory detection effect of the present invention.

[0042] Figure 5 This is a comparison chart of the incomplete trajectory detection efficiency of the present invention.

[0043] Figure 6 This is a system architecture diagram of the present invention. DETAILED DESCRIPTION

[0044] The present invention is further described in detail below in conjunction with embodiments and drawings, but the embodiments of the present invention are not limited thereto.

[0045] Example 1

[0046] like Figure 1 As shown, this embodiment discloses a method for detecting incomplete trajectories of business process event logs. First, a standardized event log is obtained; second, the obtained standardized event log is divided into a training set and a test set; then, the training set is feature extracted according to the last event and the frequency of event occurrence, and encoded; then, the encoded feature values ​​are input into a CatBoost machine learning model for iterative training, and the model parameters are optimized. After reaching a set number of iterations, an optimal CatBoost machine learning model is obtained; finally, the test set is input into the optimal machine learning model for verification to obtain the incomplete trajectories existing in the data set; the specific process is as follows:

[0047] 1) Obtain event logs through the 4TU platform, delete redundant attributes in the event logs, and only keep the traces, events, and event timestamps to form a standardized event log. The relevant business processes are recorded in this standardized event log; 8 event logs are selected, namely Env_permit, BPIC_2012, BPIC_13_i, BPIC_15_1 / 2 / 3 / 4 / 5. All event logs can be obtained through the 4TU platform. The detailed information of the selected 8 public event logs is shown in Table 1.

[0048] Table 1 Basic information statistics of event log dataset

[0049]

[0050] 2) According to the standardized event log obtained in step 1), the trajectories in the standardized event log are extracted and a data set is constructed, including the following steps:

[0051] 2.1) Traverse the event log and extract the trajectory when the trajectory identifier is accessed;

[0052] 2.2) Establishing a set E, inputting the trajectories extracted in step 2.1) into the set E to establish a data set;

[0053] 2.3) The data set obtained in step 2.2) is divided into 70% as the training set and the remaining 30% as the test set.

[0054] 3) Extract features from the training set obtained in step 2) according to the last event and the frequency of event occurrence, as follows:

[0055] Last event: traverse each track, and when the last event identifier is accessed, extract the event and record it in the end event list;

[0056] Event frequency: traverse each trajectory and count the events in the trajectory. When the same event is accessed, the counter is increased by 1. After each trajectory is traversed, the maximum value of the counter is recorded in the event list.

[0057] 4) Perform one-hot encoding on the features extracted in step 3) and update them to the training set so that the encoded feature values ​​can represent this feature, as follows:

[0058] The one-hot encoding method is used to encode the last event extracted in step 3) and the event frequency feature, and the encoded features are updated to the training set processed in step 2). The one-hot encoding is to map the categorical variables to integer values ​​and represent each value in the form of a binary vector.

[0059] 5) Input the last event encoded in step 4) and the event frequency feature value into the CatBoost machine learning model for iterative training. During the training process, the hyperparameters of the CatBoost machine learning model are continuously adjusted and optimized. After reaching the set number of iterations, the optimal CatBoost machine learning model with different features is obtained. The specific process is as follows:

[0060] 5.1) Extract the last event and the frequency of event occurrence in the training set, and input the extracted feature information into Set1 and Set2 respectively;

[0061] Given trajectory<a,b,a,c> and trajectory<a,c,b,d> , where a, b, c, d are events, then the trajectory<a,b,a,c> and trajectory<a,c,b,d> The mapping results of the last event and event frequency are as follows: Figure 2 As shown in (a) and (b).

[0062] 5.2) If Figure 3 As shown, the feature set and training set constructed in step 5.1) are input into the CatBoost machine learning model for iterative training, that is, firstly, an input sample and a target value are generated for each prefix length of the trajectory, where the input sample represents the current state of the process in the model, and the target value is a binary category label. When the current trajectory is detected as an incomplete trajectory, the target sample is "true", otherwise it is "false". Once the input sample is mapped into the feature space, it can be fed back into the model together with the target sample.

[0063] 5.3) Evaluate the model quality and adjust the model parameters based on the area under the curve (AUC) obtained from model training. After reaching the set number of iterations, the optimal CatBoost machine learning model with different feature categories is obtained.

[0064] Among them, the above-mentioned CatBoost machine learning model is a gradient boosting decision tree based on a symmetric decision tree as a learner with fewer parameters, support for categorical variables and high accuracy. It can efficiently and reasonably process categorical features, and the processing effect is better than most other machine learning algorithms. This is mainly because the CatBoost machine learning model uses the same segmentation standard for the entire tree level during the tree construction process, and each tree is built based on the residual of the previous tree, which can solve the problem caused by gradient deviation and speed up training time.

[0065] 6) Input the test set into the optimal CatBoost machine learning model with different features obtained in step 5) to detect the incomplete trajectories in the data set with different features.

[0066] In order to further verify the effectiveness and superiority of the above-mentioned business process event log incomplete trajectory detection method in this embodiment in business process management tasks, it will be compared with other detection algorithms, and the specific situation is as follows:

[0067] The experiment uses the area under the curve (AUC) as the evaluation index of each incomplete trajectory detection method. It mainly judges the accuracy of incomplete trajectory detection by calculating the area under the curve of the curve composed of the number of correctly detected samples and the number of incorrectly detected samples. If the AUC value is low, it means that the detected incomplete trajectory is more accurate.

[0068] In order to obtain the optimal model in the incomplete trajectory detection task, the extreme gradient boosting (eXtremeGradient Boosting, XGBoost) model is used as a baseline method for comparison.

[0069] Perform basic parameter adjustments on different models, and adjust the parameters as follows:

[0070] a. Learning rate: {0.01, 0.03}

[0071] b. Iterations: 100

[0072] c. Optimization algorithm: Adam

[0073] To further illustrate the advantages of the method of the present invention in terms of the accuracy of incomplete trajectory detection, the eight data sets are randomly cut according to the proportions of 0%, 10%, 20% and 30%, where 0% is the original log data. Figure 4 The AUC values ​​of 8 different data sets after the method of the present invention and the baseline method are optimized after parameter adjustment are shown. The specific analysis is as follows:

[0074] Depend on Figure 4 It can be seen that the AUC score results on the datasets containing 0% and 30% incomplete trajectories generated by random cutting show that the detection performance of the proposed method on the eight datasets is always better than that of the baseline method;

[0075] On the dataset containing 10% incomplete trajectories, the proposed method can achieve good performance on the other five datasets except that the AUC results of the BPIC_15_1, BPIC_15_2 and BPIC_15_3 datasets are lower than the baseline method;

[0076] On the datasets containing 20% ​​incomplete trajectories, the proposed method can achieve higher accuracy on the other six datasets, except that the AUC results of the Env_permit and BPIC_15_3 datasets are lower than those of the baseline method.

[0077] The overall detection effect of the method of the present invention is better than that of the baseline method, and the detection accuracy is improved by 5.6% compared with the baseline method. Therefore, it can be considered that the method of the present invention is more suitable for the task of incomplete trajectory detection than the baseline method.

[0078] In order to verify the effect of the method of the present invention on the detection time, Figure 5 The time performance of the proposed method and the baseline method on 8 different datasets is shown. Figure 5 The results show that the time performance of the method of the present invention is improved by 4.5 times, which greatly reduces the detection time of incomplete trajectories. Therefore, considering the performance of the method of the present invention in terms of detection accuracy and time performance, it is shown that the use of complex feature information and classification based on CatBoost can improve the detection and classification effect of incomplete trajectories in business processes.

[0079] Example 2

[0080] like Figure 6 As shown, this embodiment discloses a business process event log incomplete track detection system, including the following functional modules:

[0081] The data preprocessing module is used to delete redundant attributes in the event log and retain only the track, event and event timestamp, thereby forming a standardized event log, which records the relevant business process;

[0082] A data generation module is used to extract and construct a trajectory data set existing in the standardized event log obtained by the data preprocessing module, and divide the data set into a training set and a test set;

[0083] Feature extraction module, used to extract the features of the last event and the frequency of event occurrence in the training set;

[0084] A feature encoding module is used to perform targeted encoding on the features in the feature extraction module;

[0085] The model training module is used to input the feature values ​​obtained by the feature encoding module into the CatBoost machine learning model for iterative training. After reaching the set number of iterations, the optimal CatBoost machine learning model is obtained;

[0086] The detection result output module is used to input the test set into the optimal CatBoost machine learning model obtained by the model training module for verification, so as to detect whether there are incomplete trajectories in the data set and output the detection results.

[0087] In summary, the present invention proposes a method and system for detecting incomplete trajectories in business process event logs. By training a model using a machine learning method to detect incomplete behaviors in event logs, the present invention can help business process managers extract useful knowledge, thereby improving the quality of existing log data modeling. It has practical promotion value and is worthy of promotion.

[0088] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.

Claims

1. A method for detecting incomplete traces of business process event logs, characterized in that: The following steps are involved: 1) Obtain a standardized event log, which records the relevant business processes, including trajectories, events, and event timestamps; 2) According to the standardized event log obtained in step 1), the trajectories existing in the standardized event log are extracted to establish a data set, and the data set is divided into a training set and a test set; 3) Extracting features from the training set obtained in step 2) according to the last event and the frequency of occurrence of the event; 4) One-hot encode the features extracted in step 3) and update them to the training set so that the encoded feature values ​​can represent the features; 5) Inputting the encoded feature values ​​in step 4) into the CatBoost machine learning model for iterative training, and optimizing the model parameters, after reaching the set number of iterations, obtaining the optimal CatBoost machine learning model; 6) Input the test set into the optimal CatBoost machine learning model obtained in step 5) for testing, and finally detect the incomplete trajectories in the data set.

2. A method for detecting incomplete traces of business process event logs according to claim 1, characterized in that: In step 1), redundant attributes in the event log are deleted, and only the trace, event, and event timestamp are retained to form a standardized event log; all event logs are obtained through the 4TU platform.

3. A method for detecting incomplete traces of business process event logs according to claim 1, characterized in that: In step 2), according to the standardized event log obtained in step 1), the trajectories in the standardized event log are extracted and a data set is constructed. The specific steps are as follows: 2.1) Traverse the event log and extract the trajectory when the trajectory identifier is accessed; 2.2) Establishing a set E, inputting the trajectories extracted in step 2.1) into the set E to establish a data set; 2.3) The data set obtained in step 2.2) is divided into 70% as the training set and the remaining 30% as the test set.

4. A method for detecting incomplete traces of business process event logs according to claim 1, characterized in that: In step 3), the training set obtained in step 2) is subjected to feature extraction according to the last event and the frequency of occurrence of the event, as follows: Last event: traverse each track, and when the last event identifier is accessed, extract the event and record it in the end event list; Event frequency: traverse each trajectory and count the events in the trajectory. When the same event is accessed, the counter is increased by 1. After each trajectory is traversed, the maximum value of the counter is recorded in the event list.

5. A method for detecting incomplete traces of business process event logs according to claim 1, characterized in that: In step 4), the one-hot encoding method is used to encode the features extracted in step 3), and the encoded features are updated to the training set processed in step 2), wherein the one-hot encoding maps the categorical variables to integer values ​​and represents each value in the form of a binary vector.

6. A method for detecting incomplete traces of business process event logs according to claim 1, characterized in that: In step 5), the last event encoded in step 4) and two different types of feature values ​​of the event frequency are input into the CatBoost machine learning model for iterative training, that is, firstly, an input sample and a target value are generated for each prefix length of the trajectory, the input sample represents the current state of the process in the model, and the target value is a binary category label. When the current trajectory is detected to be an incomplete trajectory, the target sample is "true", otherwise it is "false". Once the input sample is mapped to the feature space, it can be fed back to the model together with the target sample; finally, the model quality is evaluated according to the area under the curve obtained by model training, and the model parameters are adjusted to obtain the optimal CatBoost machine learning model with different characteristics.

7. A method for detecting incomplete traces of business process event logs according to claim 1, characterized in that: In step 6), the test set is input into the optimal CatBoost machine learning model with different features obtained in step 5) to detect incomplete trajectories existing in the data set under different features.

8. A business process event log incomplete trajectory detection system, characterized in that: include: The data preprocessing module is used to delete redundant attributes in the event log and retain only the track, event and event timestamp, thereby forming a standardized event log, which records the relevant business processes; A data generation module is used to extract and construct a trajectory data set existing in the standardized event log obtained by the data preprocessing module, and divide the data set into a training set and a test set; Feature extraction module, used to extract the features of the last event and the frequency of event occurrence in the training set; A feature encoding module is used to perform targeted encoding on the features in the feature extraction module; The model training module is used to input the feature values ​​obtained by the feature encoding module into the CatBoost machine learning model for iterative training. After reaching the set number of iterations, the optimal CatBoost machine learning model is obtained; The detection result output module is used to input the test set into the optimal CatBoost machine learning model obtained by the model training module for verification, so as to detect whether there are incomplete trajectories in the data set and output the detection results.

Images