SOFA and multi-machine learning model-based military vocational education high-risk student prediction method
By introducing SOFA and multiple machine learning models into military vocational education, the problem of traditional assessment methods failing to identify high-risk students has been solved. Real-time monitoring and immediate feedback of students' learning dynamics have been achieved, improving teaching quality and student participation, and supporting the development of personalized teaching plans.
Patent Information
- Application Number
- CN202510454728.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-11
- Publication Date
- 2025-08-12
AI Technical Summary
In the context of massively open online military courses, traditional teaching and assessment methods struggle to identify and support high-risk students in a timely manner, resulting in high student failure and dropout rates. There is a lack of effective parameters and teaching models for identifying high-risk students.
The military vocational education teaching method based on SOFA and multiple machine learning models is adopted. Through data collection, processing and evaluation modules, combined with five theoretical and practical topics, the learning progress of students is monitored in real time and immediate feedback is provided. SMOTE technology is used to process imbalanced datasets and a voting set classifier is selected to predict high-risk students.
It enables real-time monitoring and immediate intervention of students' learning progress, improves learning motivation and teaching quality, reduces the risk of students failing and giving up, and provides personalized teaching support.
Smart Images
Figure CN120471204A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of prediction methods, and more specifically, to a prediction method for high-risk students in military vocational education based on SOFA and multiple machine learning models. Background Art
[0002] Amid the rapid development of military vocational education, Massive Open Military Online Courses (MOMCs), with their unique openness and flexibility, offer unprecedented learning opportunities to military students. Within MOMCs, the student population is large, while military education resources are relatively concentrated. Traditional teaching assessment methods, such as final assessments, struggle to provide comprehensive coverage and timely responses. While they can assess students' overall learning outcomes, their lag makes it difficult to provide timely support when students encounter learning difficulties. Courses are highly theoretical and demanding in practical application, resulting in relatively high student failure and dropout rates. Promptly identifying and monitoring high-risk students at risk of online learning failure is a key research topic for improving MOMC teaching effectiveness.
[0003] In formative assessments on massive online open courses (MOOCs), clickers and pre-course assessment scores are crucial parameters for identifying high-risk learners. Recent research has highlighted the importance of integrating online quizzes and assignments into lightweight online formative assessments (SOFA). By capturing performance metrics such as learner engagement, task completion, and simulation drills in MOOCs, it is possible to identify high-risk learners who may face learning challenges or drop out. This allows for the provision of immediate feedback to students in a cost-effective and efficient manner, promoting continuous optimization of the learning process. While there is no universally accepted model for identification algorithms, many studies have incorporated machine learning algorithms, such as logistic regression (LR), Bayesian classifiers, k-nearest neighbors (kNN), and random forests (RF). However, the success of these algorithms often depends on multiple factors, including data size, data type, and instructional methodology.
[0004] Currently, no research has demonstrated the effectiveness of lightweight online formative assessments in MOMCs for early identification of high-risk students. Furthermore, there is a lack of universal strategies or instructional models for embedding high-risk student identification parameters into military vocational education. The importance of SOFA as a key assessment tool is being addressed, and how it can be integrated into personalized learning path planning, adaptive military education resource delivery, and immediate feedback mechanisms to support student self-directed learning and instructor-directed guidance is crucial. Furthermore, it is necessary to design military vocational education instructional models that enable instructors to monitor student learning dynamics in real time, promptly identify and address potential learning issues, effectively reduce student failure and dropout risks, and enhance overall teaching effectiveness.
[0005] To address the above problems, this study focuses on the identification of high-risk students in these core courses. Through the effective application of SOFA in MOMCs, its positive role in improving the quality of military vocational education is verified, and a military vocational education teaching model suitable for the MOMCs environment is proposed. Summary of the Invention
[0006] (1) Technical problems solved
[0007] In response to the problems existing in the prior art, the present invention provides a method for predicting high-risk students in military vocational education based on SOFA and multiple machine learning models to solve the technical problems mentioned in the background technology.
[0008] (2) Technical solution
[0009] To achieve the above-mentioned objectives, the present invention provides the following technical solutions: a method for predicting high-risk students in military vocational education based on SOFA and multiple machine learning models, comprising a military vocational education teaching model module, a data collection module, a data processing module and a model evaluation module, wherein the data collection module, the data processing module and the model evaluation module are respectively connected to the military vocational education teaching model module.
[0010] The present invention further provides that the military vocational education teaching model module is divided into five theoretical topics and five practical topics. The theoretical and practical topics are well-organized and complementary, and they are carried out simultaneously with projects or problem solving. The practical topics include small problem-based assignments, which are adjusted in time in the laboratory. Since projects or problem solving are used as learning tools to expose different theoretical concepts, it is necessary to ensure that the project phase or problem-solving topic is completely consistent with the theoretical topic provided. To facilitate students' understanding, a scaffolded guidance course is also provided, showing the complete common teaching design adopted by the three courses. Three green blocks represent one week, consisting of three one-hour lectures, while one red block represents one week. The class students are divided into several groups. Both lectures and practicals are conducted in a group-based environment to promote peer and collaborative learning. The collaborative learning approach of "learning together" and "learning individually" is most appropriate for the current situation, which expects individuals to grow across the three dimensions of cooperation, competition, and individualism. Assessments for this course are divided into three categories: formative, summative, and project- or problem-based. Formative assessments are embedded throughout the course with multiple fixed-time SOFAs. The revised course evaluation schedule and score weightings are shown in Table 1. The frequency of SOFAs for all three courses is fixed at five times. SOFAs consist of multi-session questionnaires designed based on military professional education requirements.
[0011] The present invention further configures the data collection module to include a dataset obtained from an online military vocational education platform. Based on the proposed teaching model, SOFA was embedded in all three courses. All courses employed the same strategy, and all instructors participated in teaching all three courses, ensuring a consistent teaching structure. This process involved two semesters, from 2022 to 2023. The first semester was used to train the model, and the second semester's data served as a test set.
[0012] The present invention further provides that the data processing module includes the SOFA data format: Since there are five consecutive online lightweight assessments, there are five input parameters. However, because the parameters need to be validated over different time periods, five different time periods were selected. The first time period considered is the third week of the course, including the first lightweight assessment data, or LWA1. The next checkpoint is in the sixth week, including two parameters, namely the first and second lightweight assessments, namely LWA1 and LWA2. This continues in week 6, with the final checkpoint including all five lightweight assessment data, namely LWA1, LWA2, LWA3, LWA4, and LWA5. The actual scores obtained in these assessments are treated as continuous input parameters and remain unchanged. Data for absentees is marked as zero. No rows are discarded during the data preprocessing stage. The identification of high-risk and safe students divides students into two categories: high-risk students are those who are at risk of failing or dropping out. Instructors determine a threshold below which students are classified as high-risk. Students with a total score below 40% are considered high-risk students, while other students are classified as safe.
[0013] The present invention is further configured such that the data processing module also includes when analyzing high-risk students and student categories, the data set is very unbalanced. Because the number of failing students in a classroom environment is usually small compared to the number of passing students, here, the initial row represents data with an unbalanced distribution. For such an unbalanced data set, the majority class dominates, and the results are usually biased due to model overfitting. In order to convert the data into a balanced data set, we use the synthetic minority oversampling technique (SMOTE) on the data set. SMOTE is an oversampling technique that generates synthetic minority samples with the help of the k-nearest neighbor (k-NN) method, where k is set to 5. After applying SMOTE, the unbalanced data is converted into a balanced data set by oversampling the minority class.
[0014] The present invention is further configured such that the model evaluation module includes a number of benchmark models that were experimented with in accordance with the literature survey on the use of machine learning models in Section 2. The balanced training dataset was input into the following machine learning models (SVM, LR, NB, RF, XGB, kNN, DT, artificial neural network (ANN)) and their evaluation results were compared
[28] . When the results were analyzed periodically, five data sets, LWA 1, LWA 1-2, LWA 1-3, LWA1-4, and LWA 1-5, were considered for all three routes. These five periodic data sets were input into the above models respectively and their results were analyzed. Numerically, eight models were tested on five different types of data for the three routes, and 120 overall comparison results were obtained.
[0015] (3) Beneficial effects
[0016] Compared with the existing technology, the present invention provides a prediction method for high-risk students in military vocational education based on SOFA and multiple machine learning models, which has the following beneficial effects:
[0017] In the current field of military vocational education, with the increasing number of students, providing high-quality education to cultivate students' critical thinking, teamwork, and complex problem-solving skills has become a major challenge. This study addresses this issue by proposing an innovative military vocational education teaching framework. Its core is to effectively embed lightweight teaching parameters (SOFA, an online formative assessment tool) into various military vocational education courses.
[0018] Through an in-depth analysis of core courses in military vocational education, such as strategic analysis, combat simulation, and military information technology, this study found that these courses not only require students to master a solid theoretical foundation but also emphasize the cultivation of practical skills and problem-solving techniques. Therefore, the teaching framework proposed in this study aims to promote a seamless transition between theoretical learning and practical application by introducing the SOFA tool.
[0019] Specifically, this study implemented real-time monitoring and immediate feedback on student learning progress by embedding SOFA into various military vocational education courses. This formative assessment method not only increased student motivation and engagement, but also helped instructors promptly identify and intervene with students experiencing learning difficulties, thereby ensuring improved overall teaching quality.
[0020] Furthermore, this study utilized machine learning models to analyze the collected SOFA data, further exploring students’ learning patterns and potential challenges. These analyses not only provided valuable feedback to the teaching team but also provided a scientific basis for subsequent course optimization and the development of personalized teaching plans.
[0021] In summary, the SOFA-based military vocational education teaching framework proposed in this study not only effectively addresses the issue of teaching quality monitoring in large-scale military vocational education but also provides strong support for cultivating high-quality, highly skilled modern military personnel. In the future, as this teaching framework continues to be refined and promoted, it is believed that it will bring further innovation and change to the field of military vocational education. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] Figure 1 A schematic diagram of the military vocational education teaching model in the prediction method of high-risk students in military vocational education based on SOFA and multiple machine learning models;
[0023] Figure 2 Schematic diagram of using SMOTE:A3 course from unbalanced dataset (first row) to balanced dataset (second row);
[0024] Figure 3 A schematic diagram comparing the ROC-AUC scores of the three course models in two different time periods;
[0025] Figure 4 Schematic diagram of the proposed machine learning model;
[0026] Figure 5 A diagram showing the accuracy comparison of the training and test datasets of the three courses at different time periods;
[0027] Figure 6 Schematic diagram of the comparison of ROC-AUC curves of test data sets for three courses in different time periods. DETAILED DESCRIPTION
[0028] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.
[0029] It should be noted that, unless otherwise specified, all technical and scientific terms used in this application have the same meaning as commonly understood by ordinary technicians in the technical field to which this application belongs.
[0030] In the present invention, unless otherwise specified, directions such as "up" and "down" are generally used with respect to the directions shown in the drawings, or with respect to the vertical, perpendicular or gravity directions; similarly, for ease of understanding and description, "left" and "right" are generally used with respect to the left and right shown in the drawings; "inside" and "outside" refer to the inside and outside relative to the outline of each component itself, but the above-mentioned directions are not used to limit the present invention.
[0031] Military Vocational Education Teaching Model
[0032] Model Introduction
[0033] The entire teaching process is divided into five theoretical topics and five practical topics. The theoretical and practical topics are well-organized and complementary, and they are carried out simultaneously with projects or problem solving. The practical topics include small problem-based assignments
[25] , which are adjusted in time in the laboratory. Since projects or problem solving are used as learning tools to expose different theoretical concepts, it is necessary to ensure that the project stage or problem solving topic is completely consistent with the theoretical topic provided. In order to provide students with better understanding, a scaffolded guidance course is also set up. Figure 1 The complete generic instructional design used in the three courses is shown. Figure 1 In the example, three green tiles represent one week, which consists of three one-hour lectures, while one red tile represents one week. The students in the class were divided into several groups. Both lectures and practicals were conducted in a group-based environment to promote peer and cooperative learning. The cooperative learning method of "learning together" and "learning alone" is most suitable for the current situation, which expects individuals to grow in the three dimensions of cooperation, competition and individualism. The assessment of this course is divided into three categories: formative, summative and project or problem-based assessment. The formative assessment is embedded in multiple fixed-time SOFAs throughout the course. The modified course evaluation plan and score weights are shown in Table 1. The frequency of SOFA is fixed at five times for all three courses. SOFA consists of MCQs, and the MCQs used are designed based on the requirements of military professional education.
[26]
[0034] Figure 1 Military Vocational Education Teaching Model
[0035] Table 1
[0036]
[0037] method
[0038] Data Collection
[0039] The dataset was obtained from a military vocational education online platform. According to the proposed teaching model, SOFA was embedded in all three courses. All courses adopted the same strategy, and all teachers participated in the teaching of all three courses, which ensured the consistency of the teaching structure. In this process, two semesters from 2022 to 2023 were included. The first semester was used to train the model, and the data of the second semester was used as a test set
[27] .
[0040] Table 2 Detailed description of the course dataset
[0041]
[0042] Data preprocessing
[0043] SOFA Data Format: Since there are five consecutive online lightweight assessments, there are five input parameters. However, because we wanted to examine the validity of the parameters over different time periods, five different time periods were selected. The first time period considered was the third week of the course, which included the first lightweight assessment data, or LWA1. The next checkpoint was in week six and included two parameters: the first and second lightweight assessments, LWA1 and LWA2, and so on. The final checkpoint included all five lightweight assessment data, LWA1, LWA2, LWA3, LWA4, and LWA5. Details of the features used in this study are shown in Table 2. The actual scores obtained in these assessments were treated as continuous input parameters without any variation. Data for absentees was marked as zero. No rows were dropped during the data preprocessing phase. The identification of at-risk and safe students categorizes students into two main groups. At-risk students are those who are at risk of failing or dropping out. Faculty will determine a cut-off score below which students are classified as at-risk, with students scoring below 40% overall being considered at-risk, while the rest of the students will be in the safe category.
[0044] When analyzing at-risk students and student categories, the dataset is highly unbalanced because the number of failing students in a classroom setting is typically smaller compared to the number of passing students. Figure 2 The distribution of high-risk and safe categories across five lightweight assessments is shown. The initial rows represent data with an imbalanced distribution. With such an imbalanced dataset, the majority class dominates, and the results are often biased by model overfitting. To transform the data into a balanced dataset, we applied the synthetic minority oversampling technique (SMOTE). SMOTE is an oversampling technique that uses the k-nearest neighbor (k-NN) method to generate synthetic minority samples. Here, k is set to 5. After applying SMOTE, the imbalanced data is transformed into a balanced dataset by oversampling the minority class. Figure 2 The second row represents the balanced data for the A3 course distribution.
[0045] Figure 2 Using SMOTE:A3 class from unbalanced dataset (first row) to balanced dataset (second row)
[0046] Model Evaluation
[0047] Based on the literature survey on the use of machine learning models in Section 2, this study conducted experiments on some benchmark models. The balanced training dataset was input into the following machine learning models (SVM, LR, NB, RF, XGB, kNN, DT, artificial neural network (ANN)), and their evaluation results were compared
[28] . When analyzing the results periodically, five data sets, LWA 1, LWA 1-2, LWA 1-3, LWA 1-4, and LWA 1-5, for all three routes were considered. These five periodic data sets were input into the above models respectively, and their results were analyzed. Numerically, eight models were tested on five different types of data for the three routes, and 120 overall comparison results were obtained.
[0048] The top five models that outperformed other models at different stages were RF, SVM, LR, NB, and XGB. Therefore, only these five models were selected for further analysis. Figure 3 The ROC-AUC curve analysis results for LWA 1 and LWA 1-5 for all three courses are shown. Here, we can easily notice the performance differences between models in different courses and stages. For example, in LWA 1 of course A1, NB outperforms other models, while in the last LWA 1-5 stages of the same course, SVM and LR perform better
[29] .
[0049] Therefore, to maintain consistency across courses and find the most efficient and effective models for all stages, we proposed an ensemble model. A voting ensemble classifier is composed of all five models: RF, SVM, LR, NB, and XGB with soft voting. With the help of this ensemble classifier, we achieved results comparable to the best model of that stage. Figure 3 The comparison results of the ensemble classifier with other models are also shown. In all stages and classes, the voting results of the ensemble classifier are comparable to the best classifier
[30] .
[0050] Therefore, the balanced training data is fed into the voting set classifier. The classifier is trained on the training dataset. After training, the unseen balanced test data is used to make the final prediction. The complete machine model architecture is shown in Figure 2. Figure 4 As shown in the figure. The entire model is implemented in Python. Based on the research question, this study conducted two analyses. One was the prediction of students' overall course grades, and the other was the data from the five stages. The data was divided into two categories: "high-risk" and "safe". The cost of classifying high-risk students into the safe category is higher than the opposite. Therefore, in addition to accuracy, other parameters such as precision, recall, and ROC-AUC were also considered. The analysis was performed using the following evaluation parameters: accuracy, ROC-AUC, precision, recall, F1-score, and Cohen's κ statistic
[31] .
[0051] Figure 3 Comparison of ROC-AUC scores of the three course models in two different time periods
[0052] Figure 4 Proposed machine learning model
[0053] result
[0054] As mentioned previously, for early identification, the data was divided into five phases, corresponding to the five evaluation phases. For each course, five different voting ensemble classifier models were trained on the corresponding course's training dataset. Validation accuracy was calculated for each course at each phase of the training set. Figure 5 The training set validation accuracy and 5% confidence interval are shown. After that, the trained model is tested on unseen test data, and the corresponding accuracy is as follows Figure 5 The accuracy and other evaluation index parameters are shown in Table 3.
[0055] Accuracy: The comparative accuracies for the A1 and A3 courses were 74.3% and 73.3%, respectively, with κ values of 0.486 and 0.465. After the first LWA 1, held in the third week of the course, accuracy was high, with moderate agreement and reliable results. In contrast, the A2 course had a lower accuracy of 62.2%, with a κ value of 0.243.
[0056] Analysis of the evaluation metrics for the test data over the entire time period showed positive results. Model accuracy and reliability quotients improved over time for each course. However, the rate of increase varied across courses. In course A1, accuracy continued to rise, reaching a peak of 91.4% after the fifth progress point, demonstrating near-perfect consistency. In contrast, in course A2, accuracy increased at a slower rate until the second phase (LWA 1-2). However, in the third phase, accuracy increased dramatically, reaching 80.6%, and peaked at 94.7% after the fifth phase. The reliability quotient was also high, indicating perfect consistency. In course A3, accuracy continued to rise, but at a slower rate compared to the other courses. After the fifth phase, it reached a maximum of 87%, demonstrating considerable consistency.
[0057] ROC-AUC value: Figure 6The ROC-AUC plots and their values for three courses across the five phases are shown. ROC-AUC values after LWA 1 ranged from 0.700 to 0.815. These values were obtained during the third week of the course. ROC-AUC values for LWA 1-2, the sixth week, ranged from 0.739 to 0.923, and so on. The final range for LWA 1-5 was between 0.957 and 0.988. This indicates highly accurate prediction results. Even after LWA 1, ROC-AUC values for courses A1 and A3 were high, at 0.804 and 0.815, respectively. These courses achieved significantly better results than course A2, which had a relatively low ROC-AUC value of 0.700.
[0058] In terms of comparative analysis, the results cannot be directly compared due to the different teaching methods and courses used in this study. The ROC-AUC values of the two comparative studies and this study are shown in Table 3. From the perspective of the range of ROC-AUC
[39] , the results of this study are comparable to those of the PI-based study. This indicates that the proposed SOFA teaching framework is comparable to the PI-based click instrument in classroom use during the third week of the course.
[0059] Table 3 Evaluation indicators of the three in five stages
[0060]
[0061]
[0062]
[0063] Figure 5 Comparison of the accuracy of the training and test datasets of the three courses at different time periods
[0064] Figure 6 Comparison of ROC-AUC curves of test data sets of three courses at different time periods
[0065] The implemented SOFA technique was deemed effective, showing satisfactory results on the Good Teaching scale, the Clear Objectives and Standards scale, the Appropriate Assessment scale, and the Generic Skills scale (except for the Appropriate Work scale). The pedagogy was considered successful, regardless of the course structure, as students supported the course as representing good teaching and appropriate support. They were very clear about the course's learning outcomes and objectives. This also demonstrated that the course helped them develop general problem-solving skills and supported lifelong learning. The assessment structure and its relevance were highly supported and appreciated by students, as its Cronbach's ɑ value exceeded 0.9, which is close to perfect.
[0066] in conclusion
[0067] In the current field of military vocational education, with the increasing number of students, providing high-quality education to cultivate students' critical thinking, teamwork, and complex problem-solving skills has become a major challenge. This study addresses this issue by proposing an innovative military vocational education teaching framework. Its core is to effectively embed lightweight teaching parameters (SOFA, an online formative assessment tool) into various military vocational education courses.
[0068] Through an in-depth analysis of core courses in military vocational education, such as strategic analysis, combat simulation, and military information technology, this study found that these courses not only require students to master a solid theoretical foundation but also emphasize the cultivation of practical skills and problem-solving techniques. Therefore, the teaching framework proposed in this study aims to promote a seamless transition between theoretical learning and practical application by introducing the SOFA tool.
[0069] Specifically, this study implemented real-time monitoring and immediate feedback on student learning progress by embedding SOFA into various military vocational education courses. This formative assessment method not only increased student motivation and engagement, but also helped instructors promptly identify and intervene with students experiencing learning difficulties, thereby ensuring improved overall teaching quality.
[0070] Furthermore, this study utilized machine learning models to analyze the collected SOFA data, further exploring students’ learning patterns and potential challenges. These analyses not only provided valuable feedback to the teaching team but also provided a scientific basis for subsequent course optimization and the development of personalized teaching plans.
[0071] In summary, the SOFA-based military vocational education teaching framework proposed in this study not only effectively addresses the issue of teaching quality monitoring in large-scale military vocational education but also provides strong support for cultivating high-quality, highly skilled modern military personnel. In the future, as this teaching framework continues to be refined and promoted, it is believed that it will bring further innovation and change to the field of military vocational education.
[0072] In all the schemes mentioned above, the connection between the two parts can be selected according to actual conditions by welding, bolt and nut connection, bolt or screw connection or other well-known connection methods, which will not be described here one by one. In the above, all fixed connections are preferably considered to be welding. Although the embodiments of the present invention have been shown and described, it can be understood by ordinary technicians in this field that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. A prediction method for high-risk students in military vocational education based on SOFA and multiple machine learning models is characterized by: It includes a military vocational education teaching model module, a data collection module, a data processing module and a model evaluation module. The data collection module, the data processing module and the model evaluation module are respectively connected to the military vocational education teaching model module.
2. The method for predicting high-risk students in military vocational education based on SOFA and multiple machine learning models according to claim 1 is characterized by: The military vocational education teaching model module is divided into five theoretical topics and five practical topics. The theoretical topics and practical topics are well-organized and complement each other. They are carried out simultaneously with projects or problem solving. The practical topics include small problem-based assignments, which are adjusted in time in the laboratory. Since projects or problem solving are used as learning tools to contact different theoretical concepts, it is necessary to ensure that the project stage or problem-solving topics are completely consistent with the theoretical topics provided. In order to enable students to better understand, a scaffolded guidance course is also set up, showing the complete common teaching design adopted by the three courses. The three green blocks represent one week, which consists of three one-hour lectures, and one The red block represents a week. The students in the class were divided into several groups. Lectures and practices were conducted in a group-based environment to promote peer and cooperative learning. The cooperative learning method of "learning together" and "learning alone" is most suitable for the current situation, which expects individuals to grow in the three dimensions of cooperation, competition and individualism. The assessment of this course is divided into three categories: formative, summative and project or problem-based assessment. The formative assessment is embedded in multiple fixed-time SOFAs throughout the course. The frequency of SOFA for the three courses is fixed at five times. SOFA consists of MCQs, and the MCQs used are designed according to the requirements of military professional education.
3. The method for predicting high-risk students in military vocational education based on SOFA and multiple machine learning models according to claim 2 is characterized by: The data processing module includes the SOFA data format: since there are five consecutive online lightweight assessments, there are five input parameters. However, since the validity of the parameters in different time periods needs to be checked, five different phased time periods are selected. The first time period considered is the third week of the course, including the first lightweight assessment data or LWA1. The next checkpoint is in the sixth week, including two parameters, namely the first and second lightweight assessments, namely LWA1 and LWA2, and so on. The last checkpoint includes all five lightweight assessment data, namely LWA1, LWA2, LWA3, LWA4 and LWA5. The actual scores obtained in these assessments are regarded as continuous input parameters without any changes. The data of absentees are marked as zero. In the data preprocessing stage, no rows are discarded. The identification of high-risk students and safe students divides students into two categories, among which high-risk students refer to students who are likely to fail or drop out. The instructor will determine a critical value. Students below this critical value are considered high-risk students. Students with a total score below 40% are considered high-risk students, while other students belong to the safe category.
4. The method for predicting high-risk students in military vocational education based on SOFA and multiple machine learning models according to claim 1 is characterized by: The data collection module includes a data set obtained from a military vocational education online platform. According to the proposed teaching model, SOFA is embedded in all three courses. All courses adopt the same strategy, and all teachers participate in the teaching of all three courses, which ensures the consistency of the teaching structure. In this process, two semesters from 2022 to 2023 are included. The first semester is used to train the model, and the data of the next semester is used as a test set.
5. The method for predicting high-risk students in military vocational education based on SOFA and multiple machine learning models according to claim 4 is characterized by: The model evaluation module includes a literature survey on the use of machine learning models in Part 2. This study experiments with some benchmark models. The balanced training dataset is input into the following machine learning models: SVM, LR, NB, RF, XGB, kNN, DT, and artificial neural network ANN, and their evaluation results are compared. When analyzing the results periodically, five data of LWA 1, LWA 1-2, LWA 1-3, LWA 1-4, and LWA 1-5 for all three routes are considered. These five periodic data are input into the above models respectively, and their results are analyzed. Numerically, eight models are tested on five different types of data for the three routes, and 120 overall comparison results are obtained.
6. The method for predicting high-risk students in military vocational education based on SOFA and multiple machine learning models according to claim 1 is characterized by: The data processing module also includes when analyzing high-risk students and student categories, the data set is very unbalanced because the number of failing students in a classroom environment is usually small compared to the number of passing students. Here, the initial row represents data with an unbalanced distribution. For such an unbalanced data set, the majority class dominates, and the results are usually biased due to model overfitting. In order to convert the data into a balanced data set, we use the synthetic minority oversampling technique SMOTE on the data set. SMOTE is an oversampling technique that uses the k-nearest neighbor k-NN method to generate synthetic minority samples. Here, k is set to 5. After applying SMOTE, the unbalanced data is converted into a balanced data set by oversampling the minority class.