Novel second classroom prediction and early warning optimization method

Abstract

The invention discloses a novel second classroom prediction and early warning optimization method, which belongs to the technical field of artificial intelligence and machine learning, and comprises the following steps: constructing a three-index system by integrating college basic information, student registration information, behavior data and second classroom activity and growth records; according to the method and the system, student individual and tribe prediction and early warning are realized by using an SVR + CNN-LSTM + Attention model, data-driven dynamic management and intelligent decision making are realized, the technical problems of systematization, quantification and prediction in traditional college second classroom management and real-time evaluation of student participation degree, growth progress and tribe activity quality are solved, data accuracy and timeliness are improved, and the method and the system have good application prospects. According to the method, the privacy compliance is ensured, efficient processing of multi-source heterogeneous data is realized, organization and management standardization and process are realized, and an output module provides a second class report card, credit and growth record early warning, personalized activity recommendation, activity index reports and multi-dimensional data visualization.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence and machine learning technology, and in particular relates to a novel method for predicting and early warning optimization in extracurricular activities. Background Technology

[0002] Against the backdrop of deepening reforms in higher education, extracurricular activities, as an important supplement to classroom learning, play a crucial role in cultivating students' comprehensive qualities, practical abilities, and mental health. However, current management of extracurricular activities relies heavily on manual recording and post-event evaluation, resulting in significant problems such as incomplete data collection and poor real-time performance. Traditional methods often focus only on student attendance or final outcomes, failing to capture dynamic data across multiple dimensions, including behavior, voice, and facial expressions, and thus unable to accurately identify changes in student participation during activities. Furthermore, the lack of quantitative standards tailored to individual differences makes it difficult for teachers to distinguish between normal fluctuations in student participation and abnormal behavior, leading to delayed warnings of academic difficulties and psychological problems, missing the optimal window for timely intervention, and severely impacting the full realization of the educational benefits of extracurricular activities.

[0003] With the advancement of educational informatization, some institutions have attempted to introduce data-driven methods for extracurricular student management. However, existing technologies still have many limitations. On the one hand, traditional predictive models, such as single regression analysis or basic neural networks, can only extract simple features from data and cannot effectively handle the temporal correlations and key information in behavioral data, resulting in low accuracy of early warnings. On the other hand, existing systems are mostly one-way early warning models, lacking a closed-loop optimization mechanism—after generating early warning information, they cannot dynamically adjust the model based on the effects of teacher intervention and subsequent student behavior data, making it difficult to adapt to the personalized development needs of different students and diverse activity scenarios. In addition, early warning feedback lacks specificity, often using a uniform notification format, failing to provide teachers with a reference for precise intervention, ultimately resulting in low intervention efficiency and failing to meet the actual needs of refined management of extracurricular activities. Summary of the Invention

[0004] The purpose of this invention is to provide a novel prediction and early warning optimization method for extracurricular activities, which solves the technical problems of systematization, quantification, and prediction in the traditional management of extracurricular activities in universities, and real-time evaluation of student participation, progress, and the quality of community activities.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A novel method for predicting and early warning in extracurricular activities includes the following steps:

[0007] Step 1: The data acquisition module collects basic information about universities, structured information about departments and majors, and student registration information through the Internet. Sensitive information is encrypted or masked. Student registration information can be updated and verified annually.

[0008] Step 2: The behavior data collection module obtains student behavior data and campus card records from the campus system and behavior registration records, summarizes and anonymizes them daily, generates behavior characteristics and time-series characteristics, and calculates the network behavior equilibrium index and anomaly indicators.

[0009] Step 3: The activity data collection module collects activity data, including extracurricular activity information, six types of growth records, and activity credit information. Activity information includes activity number, category, time, organizer, and participation record. The six types of growth records include ideology, academics, social practice, volunteer service, cultural and sports qualities, and innovation and entrepreneurship. Activity credit information includes type credits and upper limit rules.

[0010] Step 4: After cleaning, encoding and dimensionality reduction of the data obtained in Steps 1 to 3, the SHAP method is used to evaluate the importance of features, and finally the feature data is obtained.

[0011] Step 5: The indicator construction module establishes three major indicator systems based on activity data and feature data, including the activity indicator system, the award application indicator system, and the individual ability indicator system;

[0012] Step 6: Construct individual student prediction models and tribal prediction models; the individual student prediction model takes feature data as input and outputs activity participation probability, growth record and credit prediction, and abnormal behavior warning; the tribal prediction model takes activity and feature data as input and outputs tribal activity index, quality level and development trend.

[0013] Step 7: The tribe management module calculates the tribe activity index daily and implements activity limit and downgrade mechanisms based on departmental and university-level thresholds; it also establishes mechanisms for tribe registration, member assessment, leader election, financial management, activity plan monitoring, file management, and rewards and punishments.

[0014] Step 8: The output module generates extracurricular transcripts, credit and growth record alerts, personalized activity recommendations, activity index reports, and multi-dimensional data visualization results.

[0015] Preferably, when the data collection module collects student registration information, it supports updating the registration information when students change majors, take a leave of absence, or resume their studies, and conducts annual verification according to the prescribed cycle.

[0016] Preferably, when performing step 2, the behavior data collection module acquires students' access time for educational content, access time for entertainment content, game behavior time, and campus card records. The behavior data is summarized daily to generate logs, and then anonymized and de-identified to form behavior features. The network behavior equilibrium index and anomaly indicators are further calculated to generate time-series behavior features that capture long-term trends.

[0017] Preferably, during step 3, the activity data acquisition module obtains information on extracurricular activities, six types of growth records, and activity credit information. The extracurricular activity information is collected based on student registration information and activity registration records, including activity number, category, time, organizer information, and activity participation records. The six types of growth records include ideology, academics, social practice, volunteer service, cultural and sports qualities, and innovation and entrepreneurship. The activity credit information includes activity type credits and upper limit rules.

[0018] Preferably, in step 4, the feature engineering module processes the data obtained in steps 1 to 3, including data cleaning, feature encoding, dimensionality reduction, and feature importance assessment. Data cleaning includes mode imputation for categorical features, median imputation for numerical features, and forward imputation for time-series features, and automatically marking outlier data to trigger manual review. Feature encoding includes vectorizing student registration information, constructing activity preference vectors, and encoding behavioral features. Dimensionality reduction includes using an autoencoder to reduce the dimensionality of the original features and using principal component analysis to reduce the dimensionality of tribal features. Feature importance assessment includes using the SHAP method.

[0019] Preferably, when performing step 6, the student individual prediction model inputs feature data, adopts an SVR+CNN-LSTM+Attention structure, and outputs activity participation probability, growth record achievement prediction, credit completion prediction, and abnormal behavior warning.

[0020] Preferably, when performing step 8, the output module generates a second classroom transcript, credit warning, growth record warning and personalized activity recommendations, and generates an activity index report, activity quality visualization and member participation status. It also supports annual weight optimization, cross-school comparison, regional ranking, indicator anomaly monitoring and multi-dimensional data visualization output.

[0021] This invention presents a novel second-classroom prediction and early warning optimization method, solving the technical problems of systematization, quantification, and prediction in traditional university second-classroom management, and real-time evaluation of student participation, growth progress, and the quality of community activities. The invention integrates basic university information, departmental and professional structured information, and student registration information through a data acquisition module, achieving complete recording of students' basic information and university information, supporting student information changes and annual verification, and improving data accuracy and timeliness. A behavioral data acquisition module obtains students' online behavior and campus card information, summarizing it daily and performing anonymization and de-identification processing to generate long-term time-series behavioral characteristics, ensuring privacy compliance and capturing student behavioral trends. An activity data acquisition module collects second-classroom activity information, six types of growth records, and activity credit information, systematically integrating activity participation, growth trajectory, and credit data to provide a comprehensive data foundation for subsequent prediction and analysis. Through data cleaning, feature encoding, dimensionality reduction, and SHAP feature importance assessment, the invention achieves efficient processing of multi-source heterogeneous data, improving the quality of model input and prediction accuracy, and supporting behavioral characteristics, activity preferences, and... The quantitative representation of individual abilities is achieved by establishing an activity indicator system, an award application indicator system, and an individual ability indicator system based on activity data and feature data. This forms a multi-dimensional evaluation system covering students' individual abilities, activity participation, and community management, enabling comprehensive quantitative assessment. A student individual prediction model using SVR+CNN-LSTM+Attention is constructed to predict activity participation probability, achievement of growth records, and credit completion, and to provide early warnings of abnormal behavior. Simultaneously, a community prediction model is built to predict community activity index, quality level, and development trend, enabling intelligent management at the organizational level. The community activity index is automatically calculated daily, and activity frequency limits and downgrade mechanisms are implemented based on departmental and university-level thresholds. It also supports community registration, member assessment, leader election, financial management, activity plan execution monitoring, file management, and reward and punishment mechanisms, achieving standardized and streamlined organizational management. The output module provides second-classroom transcripts, credit and growth record warnings, personalized activity recommendations, activity index reports, and multi-dimensional data visualization. It supports annual weight optimization, cross-university comparison, regional ranking, and indicator anomaly monitoring, achieving visualized and intelligent decision support. Attached Figure Description

[0022] Figure 1 This is a flowchart of the present invention;

[0023] Figure 2 This is the system architecture diagram of the present invention. Detailed Implementation

[0024] Depend on Figures 1-2 The novel second-classroom prediction and early warning optimization method shown includes the following steps:

[0025] Step 1: The data acquisition module collects basic information about universities, structured information about departments and majors, and student registration information through the Internet. Sensitive information is encrypted or masked. Student registration information can be updated and verified annually.

[0026] When collecting student registration information, the data collection module supports updating registration information when students change majors, take a leave of absence, or resume their studies, and conducts annual verification according to the prescribed cycle.

[0027] In this embodiment, the student and university information collected is used only for the analysis and prediction of extracurricular activities. All personal identification information has been anonymized, desensitized, or encrypted. It does not involve sensitive information that can directly identify an individual and is only used for statistical analysis and model training. It strictly follows data protection regulations, and all student-related information is registered, filled in, or uploaded by the students themselves.

[0028] In this embodiment, administrative information such as the name of the university, its nature of operation, establishment time, and competent authority is collected during application to establish a basic information database of universities. This database can be stored using a relational database like MySQL, with the national standard code of the university as the primary key.

[0029] Statistics on the number of students in universities, including the number of full-time undergraduate students, graduate students, international students, and the total number of faculty and staff, are compiled. The data is updated at the beginning of each semester and imported in batches using Excel spreadsheets.

[0030] Universities are categorized into seven types: finance and economics, science and engineering, comprehensive, ethnic minority, arts and law, agriculture and forestry, and teacher training, coded from CL-01 to CL-07.

[0031] The levels of higher education institutions are defined as either junior colleges or undergraduate colleges. Undergraduate colleges are further subdivided into "Double First-Class" universities and regular undergraduate universities. The data comes from the Ministry of Education's annual statistical bulletin.

[0032] Record the administrative division information of the universities, accurate to the district and county level, using the latest administrative division code from the National Bureau of Statistics, such as Haidian District of Beijing as xxxxxx.

[0033] Collect the geographical coordinates of universities in batches using a third-party map API, retaining 6 decimal places, such as xxx.xxxxxx, xx.xxxxxx.

[0034] Establish a mechanism for updating information in higher education institutions. Data updates for the previous academic year must be completed by September 15th of each academic year. Data verification rules must be set, and the growth rate of student enrollment in higher education institutions must not exceed 20%; otherwise, manual review will be triggered.

[0035] For missing information about universities, a common mean-filling method can be used. For example, if a university's type is missing, it can be classified as a science and engineering university if the proportion of engineering majors is greater than 60%.

[0036] Design information quality assessment indicators for universities, including completeness rate ≥95%, accuracy rate ≥98%, consistency, matching of administrative division codes with geographical locations, and generating quality reports monthly.

[0037] To achieve a visual display of university information, ECharts is used to create a heat map of the distribution of universities across the country, with colors assigned by type and support for zooming and clicking to view details.

[0038] For example:

[0039] The calculation scope of the activity index for finance and economics colleges is limited to extracurricular activities held by finance and economics colleges. The data comes from records in the activity information table with the college type field as CL-01, and also includes activity indicators such as the average number of activities and the average number of activities participated in per person.

[0040] The activity index for science and engineering universities is calculated using a weighted average method, with the weights allocated as follows: 40% for academic activities, 30% for skills and specialties activities, and 30% for innovation and entrepreneurship activities. The basic data is the activity participation data from the past three semesters, which is also correlated with the average activity score and the average activity participation score per person.

[0041] The comprehensive university activity index comprises 12 secondary indicators, covering all activity types including ideological growth and volunteer service. The indicator weights are determined using the entropy weight method, and the calculation formula is as follows: Among them W k S represents the weight of the k-th indicator. k To standardize the scoring, the indicator system has added average activity category and average registration success rate.

[0042] The National Colleges and Universities Activity Index specifically includes the participation rate of ethnic cultural activities, defined as NCI = number of participants in ethnic activities / total number of participants × 100%, with a threshold set at ≥15%. The average attendance rate and average assignment submission rate are also statistically analyzed.

[0043] The activity index for liberal arts colleges focuses on the quality of social practice activities, and is calculated by multiplying the average activity score by the number of participants. The average rating is N, where N is the number of participants. A new average activity positive review rate metric has been added. The number of valid positive reviews for an activity is defined as an activity rated three stars or higher. The activity positive review rate = number of valid positive reviews for an activity / number of participants in the activity.

[0044] The activity index for agricultural and forestry colleges introduces a seasonal adjustment factor, increasing the weight of spring and autumn by 20% and decreasing it by 10% in summer and winter to adapt to the characteristics of agricultural seasons and simultaneously calculate the average activity leave rate.

[0045] The activity index for teacher training colleges includes specific indicators for teacher skills activities, which statistically analyze the frequency and quality scores of participation in activities such as microteaching and lesson plan design, with a weighting of no less than 25%, and is associated with the average ability value in the individual ability indicators.

[0046] The activity index for college types is updated monthly, using a sliding window technique to retain data from the most recent 12 months, and employing exponential smoothing to eliminate short-term fluctuations with a smoothing coefficient α = 0.3.

[0047] The activity index warning thresholds for different types of universities are designed. A score below 65 for science and engineering universities triggers a warning, while a score below 60 for comprehensive universities triggers a warning. The warning information is then pushed to the activity management section of the university's academic affairs office.

[0048] Establish a comparative analysis mechanism for activity indices of different types of universities, generate a ranking table of indices for seven types of universities every quarter, and mark the year-on-year change rate to provide data support for cross-university activities and cooperation.

[0049] The activity index for vocational colleges focuses on skills training activities, including indicators such as vocational qualification certification training and practical skills competitions. The participation rate target is set at ≥60%, and the average number of award applications is also included in the award application indicators.

[0050] The undergraduate activity index distinguishes between Double First-Class universities and regular undergraduate universities. Double First-Class universities have added an indicator for international exchange activities, with a weight of 15%, while regular undergraduate universities have added an indicator for employment guidance activities, with a weight of 20%. A new indicator is the average success rate of award applications per person, which is the number of award applications that have passed the review divided by the total number of awards applied for.

[0051] The activity index calculation for vocational colleges adopts the base adjustment method, which standardizes the activity duration to 120 minutes per activity, and activities shorter than 60 minutes are counted as 0.5 activities.

[0052] The undergraduate college activity index introduces a teacher-student ratio correction factor. When the teacher-student ratio of the activity organization is less than 1:30, the index is reduced by 5% to ensure the quality of the activity.

[0053] A conversion model for activity indices at the higher education level was established. The junior college index was converted into the undergraduate standard score using the formula BU = 1.2 × CC + 15, which facilitates cross-level comparison.

[0054] The activity index at the university level is analyzed weekly to predict the index changes over the next four weeks using linear regression. An early warning is issued when the predicted value falls below a threshold.

[0055] Sensitivity analysis of the activity index at the university level identified the three most influential indicators as participation rate, activity type diversity, and score, which were then identified as key monitoring targets. Activity type diversity was correlated with the average activity classification index.

[0056] The design incorporates a dashboard for activity indices at the university level, displaying the current index, target value, and warning status in real time. It also supports querying historical data by semester and academic year, with data dimensions covering activity indicators, award application indicators, and individual ability indicators.

[0057] Establish a mechanism linking university-level activity index with resource allocation. For every 5 points increase in the index, activity funding will increase by 3% in the following quarter, creating a positive incentive.

[0058] The process for handling abnormal fluctuations in the activity index at the university level automatically triggers data verification when the daily index change exceeds ±10%, checking for errors in the activity data entry.

[0059] The university type encoding is converted into a one-hot vector, which is used as one of the input features of the SVR model, with a dimension of 7.

[0060] The number of students in higher education institutions is transformed logarithmically, i.e., ln(N+1), and then used as a feature to eliminate the influence of dimensions, where N is the total number of full-time students.

[0061] The university level serves as the basis for stratified training of the model, with three sub-models trained separately for junior colleges, regular undergraduate programs, and "Double First-Class" universities, thereby improving prediction accuracy.

[0062] Information on the administrative divisions where universities are located is used to construct regional activity demand models. Universities in the same administrative division share regional characteristics, such as local policies and cultural features.

[0063] The time series of the activity index of college types is used as one of the inputs to the CNN-LSTM-ATTENTION model, with a time step of 16, corresponding to 4 months.

[0064] Gradient boosting tree was used to calculate the importance of university features, and the top 5 features, namely type, level, number of students, district, and establishment time, were selected to participate in model training.

[0065] Establish association rules between university information and activity needs, such as finance and economics universities having a high proportion of demand for innovation and entrepreneurship activities, with a confidence level of 0.78.

[0066] The missing values ​​of university information were handled using the KNN imputation method, where K=5, and the average of the 5 universities with the highest similarity was selected as the imputation value.

[0067] Principal component analysis was performed on the information from universities, reducing the dimensionality of the original 8 features to 3 principal components with a cumulative contribution rate of ≥85%, thus reducing the complexity of the model.

[0068] The standardization of information features in higher education institutions adopts the interval scaling method. Map the eigenvalues ​​to the interval [0,1].

[0069] Sensitive information in universities is stored using AES-256 encryption, such as faculty and staff salaries and students' personal contact information. The encryption key is changed every quarter.

[0070] A university information access control matrix is ​​established, divided into three levels: administrator, analyst, and visitor. Administrators have full permissions, analysts have read-only permissions, and visitors have access to statistical data.

[0071] Universities use the HTTPS protocol for information transmission, with a transmission timeout of 30 seconds. If the transmission fails, it will automatically retry, up to 3 times.

[0072] Regularly perform information anonymization processing for universities. Data used for model training should have fields that can identify individuals removed, such as name and student ID.

[0073] Establish a university information operation log to record all data query, modification, and deletion operations. The log should be retained for 2 years.

[0074] The university information database is backed up daily with incremental backups and weekly with full backups. The backup files are stored on a remote server.

[0075] An intrusion detection system is used to monitor abnormal access to university information. If the same IP address fails to log in ≥5 times within 1 hour, the IP address will be automatically blocked for 24 hours.

[0076] The university's information system has passed the Level 2 certification of the National Information Security Protection System and undergoes a security assessment annually.

[0077] Develop emergency response plans for information leaks in universities, clarify emergency response procedures, responsible persons and contact information, and organize emergency drills every six months.

[0078] Data usage agreements were signed with universities, clearly defining the data's purpose as limited to predicting extracurricular activities, the usage period as three years, confidentiality obligations, and automatic deletion of the data upon expiration of the agreement.

[0079] The scope of student personal information collection is limited to 10 basic fields: name, student ID, university code, department, major, grade, class, year of enrollment, gender, and ethnicity. Student personal information is obtained through voluntary registration by students via the client application (APP).

[0080] The student ID number encoding rule is: 4 digits of the year of enrollment + 2 digits of the department code + 2 digits of the major code + 2 digits of the class number + 2 digits of the individual number. For example, xxxxxxxxxx represents x grade x department x major x class x number.

[0081] The university code for each student uses the national standard code, such as Peking University as xxxxx.

[0082] Grade information is dynamically calculated using the year of enrollment plus the length of study. For example, the length of study for undergraduate students in the xxxx class is x years. In xxxx, they are in their third year and the code is xx.

[0083] Student personal information is stored in a distributed manner, with basic information stored in a relational database and behavioral data stored in MongoDB, linked by student ID.

[0084] Establish a student information change mechanism. When a student changes majors, takes a leave of absence, or resumes studies, the system information must be updated within 72 hours and synchronized with historical data.

[0085] Student personal information is uniquely verified, with student ID numbers and national ID card numbers being unique. National ID card numbers are stored in encrypted form to prevent duplicate entries.

[0086] The timeframe for collecting students' personal information is to complete the initial collection within 15 days of enrollment, and to conduct information verification and updates every September.

[0087] Design a student personal information entry interface and use form validation technology to ensure that the field format is correct, such as the year of enrollment being a 4-digit number.

[0088] Sensitive personal information of students is displayed in a masked format. For example, the ID number can only be viewed in its entirety by the administrator. The format is **************.

[0089] Step 2: The behavior data collection module obtains student behavior data and campus card records from the campus system and behavior registration records, summarizes and anonymizes them daily, generates behavior characteristics and time-series characteristics, and calculates the network behavior equilibrium index and anomaly indicators.

[0090] In step 2, the behavior data collection module acquires students' access time for educational, entertainment, and gaming activities, as well as campus card records. The behavior data is summarized daily to generate logs, and then anonymized and de-identified to form behavior features. The network behavior equilibrium index and anomaly indicators are further calculated to generate time-series behavior features that capture long-term trends.

[0091] Step 3: The activity data collection module collects activity data, including extracurricular activity information, six types of growth records, and activity credit information. Activity information includes activity number, category, time, organizer, and participation record. The six types of growth records include ideology, academics, social practice, volunteer service, cultural and sports qualities, and innovation and entrepreneurship. Activity credit information includes type credits and upper limit rules.

[0092] When performing step 3, the activity data collection module acquires information on extracurricular activities, six types of growth records, and activity credit information. The extracurricular activity information is collected based on student registration information and activity registration records, including activity number, category, time, organizer information, and activity participation records. The six types of growth records include ideology, academics, social practice, volunteer service, cultural and sports qualities, and innovation and entrepreneurship. The activity credit information includes activity type credits and upper limit rules.

[0093] In this embodiment, the extracurricular activity numbers that students participate in are in UUID format and correspond one-to-one with the activity numbers in the activity information table.

[0094] When counting the number of times students participate in activities, records of duplicate participation in the same activity are filtered out. The activity number and student ID are used as a joint primary key to remove duplicates. The data is then used to calculate the average number of activities participated in per student.

[0095] The number of student growth records is counted by academic year and is divided into six categories: ideological growth, academic development, social practice, volunteer service, cultural and sports qualities, and innovation and entrepreneurship. Each category is counted separately and linked to personal growth record indicators, which are the fluctuations in personal ability scores recorded within a specified period.

[0096] The rules for calculating extracurricular credits are as follows: 1 credit per academic activity, 0.5 credits per arts and sports activity, and 0.5 credits for every 2 hours of volunteer work, with a maximum of 8 credits per academic year.

[0097] The roles in the activity courses are divided into three categories: organizer, facilitator, and participant. Each role is assigned a different weight to participate in the overall score, with weights of O: 2.0, A: 1.5, and P: 1.0. The score data is included in the calculation of the average activity score.

[0098] Create a heatmap of student activity participation, with the horizontal axis representing weeks 1-18 and the vertical axis representing activity type. The color intensity indicates the frequency of participation, visually demonstrating the participation patterns.

[0099] The continuity index of student activity participation is calculated as C = max{k|ti + k - 1 - ti + 1 = k, k ≥ 1}, where ti is the number of weeks of participation.

[0100] The participation rate for cross-semester activities is calculated as the number of participants in two consecutive semesters / the total number of participants × 100%, with a target value of ≥40%.

[0101] The student activity participation preference vector is constructed as P = [p1, p2, ..., p7], where pk is the proportion of participation in the k-th type of activity, satisfying the following condition: The data is used to analyze average activity classification.

[0102] Design a student activity participation leaderboard, ranking students based on three dimensions: total number of participations, number of credits earned, and diversity of activity types. The leaderboard will be updated every Monday, with the diversity of activity types corresponding to the average activity category index.

[0103] The growth record count is calculated on a calendar year basis. The data for that year is frozen at 24:00 on December 31st each year, and the count restarts on January 1st of the following year.

[0104] The growth record review process is as follows: submission by the activity leader → review by the department administrator → final review by the school administrator. Each step takes 2 working days to process, and the review results affect the success rate of award applications.

[0105] The rule linking second classroom credits with first classroom credits is that achieving ≥16 second classroom credits is a prerequisite for the first classroom credit graduation qualification review.

[0106] The credit recognition objection handling mechanism allows students to submit an appeal within 7 working days if they disagree with the credit recognition results, which will be reviewed by the school's Second Classroom Management Center.

[0107] The data visualization of growth records uses radar charts to show the distribution of six types of growth records, line charts to show the trend of changes during the semester, and supports exporting PDF reports. The charts include average ability values, which are ability scores / number of activities.

[0108] Establish a growth record early warning mechanism. When the number of a certain type of growth record is 0 for two consecutive months, recommend activities of that type to the student.

[0109] The rules for converting extracurricular credits are as follows: winning awards in national-level competitions can be converted into innovation and entrepreneurship credits at a ratio of 1:2, with 1 competition credit equal to 2 extracurricular credits. Competition data is included in the statistics of the average number of award applications.

[0110] Upon graduation, students' extracurricular credits and growth records are automatically generated into an "Extracurricular Transcript," which serves as one of the materials for job recommendations. The transcript contains data related to individual ability indicators.

[0111] The data archived for students' growth records will be kept for 5 years after graduation. After that period, the data will be automatically converted into historical archives and compressed for storage to save space.

[0112] Develop a credit warning app function to send a credit warning notification when a student needs to obtain ≥3 credits for the remaining semester and their current progress is <50%.

[0113] The statistics on students' online access time to ideological and political content are calculated using URL matching, matching against the Ministry of Education's whitelist of ideological and political education websites.

[0114] The duration of access to entertainment video types is identified by the app package name, such as Tencent Video and iQiyi. The statistics are calculated in hours and are rounded to one decimal place.

[0115] The game type access count excludes educational games, such as programming learning games, and only counts purely entertaining games. The count is reset daily.

[0116] Access time for educational content includes three categories: online courses, academic databases, and e-books, accounting for 40%, 35%, and 25% respectively.

[0117] Student online behavior information is summarized daily and stored in JSON format: {"date":"2023-10-01","Ideological and Political Education":2.5,"Entertainment":1.8,"Games":0,"Education":3.2}.

[0118] The Network Behavior Equilibrium Index is calculated as NEI = 0.4 × E + 0.3 × P + 0.2 × S - 0.1 × G, where E is the duration of education, P is the duration of ideological and political education, S is the duration of social interaction, and G is the duration of gaming.

[0119] The detection of abnormal student online behavior is defined as follows: if the daily gaming time exceeds 5 hours or the entertainment time exceeds the educational time for 3 consecutive days, the behavior is considered abnormal.

[0120] The extraction of network behavior time series features uses a sliding window to calculate the mean, variance, and peak value of each type of duration. The window size is 7 days, forming a 21-dimensional feature vector.

[0121] A correlation model between online behavior and activity participation was established, revealing a pattern that for every additional hour spent on educational activities, the probability of participation in academic activities increased by 8%, which was correlated with the average registration success rate.

[0122] Student online behavior data is anonymized by removing identifiable information such as IP address and MAC address, retaining only behavioral characteristics and student ID, which is then encrypted.

[0123] When student grade information is used as input features of the model, it is converted into the number of years since the year of enrollment, such as the 2021 class being 3 years, and then normalized to [0,1] using min-max.

[0124] Professional information is converted into 10-dimensional vectors using the Word2Vec model, and the training corpus is the 2023 version of the National University Professional Name Database.

[0125] The student activity type preference vector P is used as one of the input features of the SVR model to construct individualized participation criteria and correlate them with the average activity classification index.

[0126] The ratio of the number of growth records to extracurricular credits is used as an indicator of learning input. The formula is R = number of growth records / credits, with an ideal value of ≥ 5 records / credit.

[0127] Step 4: After cleaning, encoding and dimensionality reduction of the data obtained in Steps 1 to 3, the SHAP method is used to evaluate the importance of features, and finally the feature data is obtained.

[0128] In step 4, the feature engineering module processes the data acquired in steps 1 to 3, including data cleaning, feature encoding, dimensionality reduction, and feature importance assessment. Data cleaning includes mode imputation for categorical features, median imputation for numerical features, and forward imputation for time-series features, and automatically marks outlier data to trigger manual review. Feature encoding includes vectorization of student registration information, construction of activity preference vectors, and encoding of behavioral features. Dimensionality reduction includes using an autoencoder to reduce the dimensionality of the original features and using principal component analysis to reduce the dimensionality of tribal features. Feature importance assessment includes using the SHAP method.

[0129] The 21-dimensional temporal features of student network behavior are input into the LSTM layer of the CNN-LSTM-ATTENTION model to capture long-term behavioral trends.

[0130] Chi-square test was used to screen student characteristics that significantly influenced activity participation (p < 0.05). Eight characteristics, including major, grade, and online behavior equilibrium index, were retained. The mean values ​​of each item in the activity indicators were calculated based on the correlation between the characteristic data and the activity indicators.

[0131] Missing values ​​in student personal information are handled by using mode imputation for categorical features, median imputation for numerical features, and forward imputation for time features. The imputation results are used to calculate individual ability indicators.

[0132] Establish a mapping relationship between individual student characteristics and activity needs. For example, the probability of STEM students needing skills-related activities is 0.62. The mapping relationship is based on the distribution of indicators such as average activity score and average activity participation score per student.

[0133] The importance of student characteristics is assessed using the SHAP value. When the SHAP value of a professional characteristic is greater than 0.2, it is considered a high-impact characteristic. The high-impact characteristics are weighted and calculated based on their correlation with indicators such as average activity category and average registration success rate.

[0134] The optimization of student personal information feature space adopts an autoencoder to compress the original 38-dimensional features to 15-dimensional features with a reconstruction error of ≤0.05. The compressed features include the core dimensions of activity indicators, award application indicators, and personal ability indicators.

[0135] The threshold for the activity index of departmental-level tribes is set at 70 points. Tribes that score below the threshold will be limited to holding no more than 2 activities per month. The index is calculated based on the average number of activities and the average attendance rate of the tribe organizations.

[0136] Step 5: The indicator construction module establishes three major indicator systems based on activity data and feature data, including the activity indicator system, the award application indicator system, and the individual ability indicator system;

[0137] This embodiment establishes an activity indicator system based on extracurricular activity information, activity preferences, and behavioral characteristics; establishes an award application indicator system based on activity data and historical award records; and establishes a personal ability indicator system based on six types of growth records, activity credits, and behavioral characteristics.

[0138] In this embodiment, the activity indicator data collection frequency is set to real-time collection, with dynamic indicators such as the average number of activities and the average number of participants per person updated every hour to ensure data timeliness.

[0139] The award application indicators are stored using blockchain technology. Once the application data is uploaded to the chain, it cannot be tampered with, ensuring the authenticity of the original data used to calculate the average award application success rate.

[0140] The calculation of the average ability value in the individual ability indicator introduces a decay factor. The activity score that is further away from the current time has a lower weight. The decay coefficient decreases by 0.1 quarterly, with a maximum decay to 0.3.

[0141] Establish an outlier identification mechanism for activity indicators. When the average activity score exceeds 120% of the full score or the average registration success rate exceeds 100%, it will be automatically marked as an anomaly and a manual review will be triggered.

[0142] The analysis of the linkage between award application indicators and activity indicators shows that when the average activity score of a certain type of activity is ≥85 points, the success rate of award applications for that activity increases by 15%, forming a positive correlation.

[0143] Personal growth record indicators are integrated with campus card data, and behavioral data such as consumption time and library visit frequency are used to help verify the authenticity of the records and improve data credibility.

[0144] The average homework submission rate in the activity indicators is linked to the recognition of extracurricular credits. For activities with a homework submission rate of less than 60%, the credits for participating students in that activity will be calculated at half the rate.

[0145] The award application indicators are set with tiered thresholds. Organizations that submit ≥30% of their applications for national-level awards will receive an additional 10 points in their annual award assessment to encourage applications for high-level awards.

[0146] Individual ability indicators are linked to students' comprehensive assessment. Students whose average ability scores rank in the top 20% will receive an additional 5 points in the extracurricular activity portion of the comprehensive assessment, thus strengthening the ability-oriented approach.

[0147] The average activity category in the activity indicators is calculated based on its matching degree with the university's major settings. Activities with a matching degree of ≥80% will have their activity scores calculated at 1.2 times the normal rate to improve the relevance of the activities.

[0148] The award application criteria implement a pre-application system. Pre-applications can be submitted 3 days before the event. The system will provide an estimated probability of approval based on the historical average success rate of award applications to assist in application decisions.

[0149] The personal growth record indicators set an annual growth threshold. Students whose growth value increases by ≥30% compared to the previous year will be awarded the title of "Growth Star" and included in the evaluation criteria.

[0150] The average positive feedback rate of an event is linked to the performance of the event organizer. Organizers with a positive feedback rate of ≥90% for three consecutive months will be awarded an "Outstanding Organizer" certificate to incentivize the improvement of event quality.

[0151] The award application indicators establish a cross-school comparison mechanism, and publish the ranking of the average award application success rate of universities in the region every quarter. Universities in the bottom 10% are required to submit improvement plans.

[0152] Personal ability index data are incorporated into the student employment recommendation algorithm. For every 1 point increase in the average ability score, the recommendation priority of high-quality enterprises increases by 2%, thereby enhancing the application value of abilities.

[0153] The average attendance rate, one of the activity metrics, is collected using facial recognition technology. The attendance data is synchronized to the system in real time, reducing data distortion caused by proxy signing and ensuring accurate calculation.

[0154] Award application quotas are linked to the professional titles of supervising teachers. For award application projects supervised by professors, the number of applications is calculated at 1.1 times the number of professors, which encourages teachers with high professional titles to participate in supervision.

[0155] The personal growth record indicators include a student self-assessment channel, allowing students to submit one self-assessment report on their abilities and growth each semester. The self-assessment results account for 15% of the total score of the personal growth record, promoting self-reflection.

[0156] An early warning threshold is set for the average absence rate in the activity indicators. For activity types with an absence rate of ≥20% for two consecutive months, the approval of that type of activity will be suspended for one month, forcing the improvement of activity quality.

[0157] The three major indicator systems (activity indicators, award application indicators, and individual ability indicators) are optimized in weight each academic year. The Delphi method is used to invite 50 experts to vote on the adjustments to ensure the adaptability of the indicator system.

[0158] Step 6: Construct individual student prediction models and tribal prediction models; the individual student prediction model takes feature data as input and outputs activity participation probability, growth record and credit prediction, and abnormal behavior warning; the tribal prediction model takes activity and feature data as input and outputs tribal activity index, quality level and development trend.

[0159] In step 6, the student individual prediction model is input with feature data and uses an SVR+CNN-LSTM+Attention structure to output activity participation probability, growth record achievement prediction, credit completion prediction, and abnormal behavior warning.

[0160] Step 7: The tribe management module calculates the tribe activity index daily and implements activity limit and downgrade mechanisms based on departmental and university-level thresholds; it also establishes mechanisms for tribe registration, member assessment, leader election, financial management, activity plan monitoring, file management, and rewards and punishments.

[0161] In this embodiment, the activity index of the school-level tribe must reach 80 points or above each quarter. If it fails to meet the standard for two consecutive quarters, it will be downgraded to the department-level tribe. The index is composed of the average activity approval rate and the average success rate of award application per person.

[0162] The data required for calculating the tribe activity index is automatically extracted from the activity information table and the participation data table. The calculation task is executed at 3:00 AM every day, and the calculation dimensions cover the average submission rate and the average absence rate of the activity indicators.

[0163] The results of the tribe activity index are synchronized to the tribe administrators and mentors via email and system notifications, so as to facilitate timely understanding of the tribe's status. The results details include the average ability value distribution in the individual ability indicators.

[0164] In this embodiment, a tribal activity index appeal mechanism can also be established based on actual usage. Tribes that disagree with the calculation results can submit appeal materials within 5 working days after the results are published. The appeal review will refer to the original data of the award application indicators.

[0165] If any abnormal data is found during the calculation of the tribal activity index, such as a negative number of participants, the system will automatically mark it and suspend the calculation, and notify the data administrator for processing. Abnormal data involves the correction of the basic statistical values ​​of the activity index.

[0166] Historical data on tribal activity index is retained for 5 years and archived at the end of each year to facilitate trend analysis and long-term evaluation. The archived data includes the change curves of activity indicators and award application indicators for each year.

[0167] The tribe activity index is directly linked to the tribe evaluation. The top 10% of tribes in the annual index will be awarded the title of "Excellent Tribe" and will be given additional activity funding. The evaluation criteria include the average diversity of activity categories and the growth rate of the number of activities participated in per person.

[0168] Differentiated index calculation rules were developed for different types of organizations. For example, student unions focused on service coverage and correlated with the average number of activities and the average registration success rate; clubs focused on the innovation of activities and correlated with the average activity score and the average activity approval rate.

[0169] The annual registration system for tribes is conducted every September. Tribes that do not register will have their eligibility for activities automatically frozen. Registration information includes the completion status of award application targets in the previous academic year.

[0170] The criteria for evaluating members of the tribe are that they must participate in no fewer than 4 activities per semester and serve no fewer than 8 hours of volunteer service. Those who fail the evaluation will be converted into probationary members, and the evaluation results will be included in their personal growth record indicators.

[0171] The tribal leader is elected through democratic elections for a one-year term, which can be renewed once. The election process is supervised by a mentor, and the leader's qualifications are linked to the average ability value in the individual's ability indicators.

[0172] The tribe's financial management system involves establishing a dedicated account, keeping detailed records of all income and expenditure, publishing monthly financial statements, accepting member supervision, and evaluating the efficiency of fund utilization in conjunction with the tribe's average number of activities and activity scores.

[0173] Tribal activity plans are to be submitted at the beginning of each semester, including the activity theme, time, budget, and expected results. If more than 30% of the activities are not carried out according to the plan, it will affect the assessment. The completion rate of the plan is related to the average achievement rate of the activity category.

[0174] The tribal archives management regulations include the establishment approval document, bylaws, member list, activity records, financial statements, etc. The archives are retained for the duration of the tribe's existence plus 5 years. The activity records must fully record all the original data of the activity indicators.

[0175] The evaluation cycle for the tribe consists of monthly self-evaluation, quarterly assessment, and annual final evaluation. The evaluation results are posted on the tribe's bulletin board for no less than 3 working days. The quarterly assessment focuses on the phased completion of the award application indicators.

[0176] The tribe reward and punishment mechanism is to award 10,000 yuan in activity funds to outstanding tribes in the year. Tribes that fail the assessment for two consecutive years will be deregistered. The reward and punishment are based on a comprehensive score that includes activity indicators and individual ability indicators.

[0177] The process for merging and splitting tribes is as follows: merging requires the consent of more than 2 / 3 of the members of both tribes, and splitting requires the consent of more than 3 / 4 of the members of the original tribe. It is also subject to approval by the school's Youth League Committee, with reference to the tribe's historical average activity score and the number of activities participated in per person.

[0178] The assessment of instructors in the community requires them to conduct no fewer than 6 guidance activities and submit no fewer than 2 guidance reports per year. The assessment results are included in the instructors' annual workload, and the guidance effectiveness is linked to the average homework submission rate and the average activity satisfaction rate among the activity indicators.

[0179] Tribal type encoding is converted into a 5-dimensional one-hot vector, which serves as one of the input features of the SVR model. The feature weights are based on the average activity classification preference of the tribe.

[0180] Tribal activity index standardized processing The index is mapped to the [0,1] interval, and the standardized benchmark includes the historical extreme values ​​of activity indicators and award application indicators.

[0181] The tribe level serves as an adjustment parameter for the model. Five-star tribes have a 20% increased weight in predicting activity participation, while one-star tribes have a 10% decreased weight. The level assessment is based on the overall improvement in individual ability indicators.

[0182] Tribe establishment year converted into a trait:

[0183] Age=log(currentyear-establishyear+1);

[0184] To reduce the impact of the time dimension, a correlation analysis was conducted between the annual characteristics and the growth trend of the average number of activities.

[0185] Tribal member size, after normalization, is used as a feature: Modeling the correlation between scale characteristics and the number of activities per person and the average activity load.

[0186] The preference vector for tribe activity types was constructed by statistically analyzing the proportion of various activities held by tribes in the past three years, forming a 7-dimensional preference vector. The preference distribution was cross-validated with the average activity classification index.

[0187] The mean and variance of the quality score of tribal historical activities are used as input features of the model to reflect the stability of tribal activities. The quality score includes the comprehensive calculation results of the average activity score and the average activity approval rate.

[0188] Tribal activity characteristics are calculated as follows This refers to the average number of activities per person per year, and the correlation analysis between activity level, average number of activities, and continuous indicators of member participation.

[0189] Principal component analysis was used to reduce the dimensionality of the tribe's multi-dimensional features, retaining the principal components with a cumulative contribution rate of 90%. The principal components contain the core information of activity indicators and award application indicators.

[0190] The cross-combination of tribal characteristics and student characteristics, such as the combination of "club tribe + science and engineering students", adjusts the predicted weight of activity participation. The combination weight takes into account the matching degree of average activity participation score and individual ability value.

[0191] Establish a tribal collaboration platform to support cross-tribal activity cooperation applications, resource sharing, and personnel allocation. The evaluation of collaboration effectiveness includes the average growth rate of the number of activities and the average increase in the registration success rate.

[0192] The scope of tribe resource sharing includes venues, materials, lecturer resources, activity experience, etc. Sharing records are included in the tribe's assessment bonus points, and the bonus points are linked to the increase in the average activity score.

[0193] The approval process for cross-tribal activities is as follows: the initiating tribe submits a cooperation plan → both instructors review it → the school's Youth League Committee approves it. The approval time limit is 5 working days. The approval criteria include the expected average diversity of activity categories and the potential for award applications.

[0194] The tribal alliance formation mechanism allows tribes of the same or complementary types to form alliances. Resources for activities within the alliance are shared preferentially. Outstanding alliances are selected annually, and the alliance's advantages are evaluated based on the average number of activities participated in per person and the average positive feedback rate of activities within the alliance.

[0195] The mechanism for sharing experiences in tribal activities involves holding monthly exchange meetings for tribal leaders and compiling outstanding activity cases into a book for all tribes to learn from. The selection criteria for these cases include outstanding performance across various activity metrics.

[0196] The construction of the tribal resource database is to classify and organize the resource information that can be shared by various tribes, including resource descriptions, available time periods, application processes, support online reservations, and conduct correlation analysis between resource utilization efficiency and average number of activities and activity completion rate.

[0197] The inter-tribal competition and cooperation balance mechanism sets up a cooperation index and a competition index, with the ratio of the two maintained at 6:4 to promote healthy development. The cooperation index includes the average submission rate of cross-tribal activities, while the competition index includes the average activity score ranking of similar activities.

[0198] Cross-campus tribal collaboration support establishes inter-campus tribal collaboration channels, unifies activity standards and data interfaces, facilitates resource sharing, and synchronously calculates the average number of participants per activity and the average coverage of activity categories across campuses.

[0199] The evaluation indicators for the effectiveness of tribal collaboration are the increase in participation rate of collaborative activities, the reduction rate of resource sharing costs, and the increase in member satisfaction. The increase in participation rate is calculated based on the baseline value of the average number of activities.

[0200] The tribal collaboration incentive policy adds 5 points to the assessment score of each tribe that successfully organizes cross-tribal activities. Outstanding cooperation cases will be awarded additional activity funds. The assessment score will be included in the award application index of the tribal activity index.

[0201] Step 8: The output module generates extracurricular transcripts, credit and growth record alerts, personalized activity recommendations, activity index reports, and multi-dimensional data visualization results.

[0202] When performing step 8, the output module generates extracurricular transcripts, credit warnings, growth record warnings, and personalized activity recommendations. It also generates activity index reports, activity quality visualizations, and member participation data. In addition, it supports annual weight optimization, cross-school comparisons, regional rankings, indicator anomaly monitoring, and multi-dimensional data visualization output.

[0203] This invention presents a novel second-classroom prediction and early warning optimization method, solving the technical problems of systematization, quantification, and prediction in traditional university second-classroom management, and real-time evaluation of student participation, growth progress, and the quality of community activities. The invention integrates basic university information, departmental and professional structured information, and student registration information through a data acquisition module, achieving complete recording of students' basic information and university information, supporting student information changes and annual verification, and improving data accuracy and timeliness. A behavioral data acquisition module obtains students' online behavior and campus card information, summarizing it daily and performing anonymization and de-identification processing to generate long-term time-series behavioral characteristics, ensuring privacy compliance and capturing student behavioral trends. An activity data acquisition module collects second-classroom activity information, six types of growth records, and activity credit information, systematically integrating activity participation, growth trajectory, and credit data to provide a comprehensive data foundation for subsequent prediction and analysis. Through data cleaning, feature encoding, dimensionality reduction, and SHAP feature importance assessment, the invention achieves efficient processing of multi-source heterogeneous data, improving the quality of model input and prediction accuracy, and supporting behavioral characteristics, activity preferences, and... The quantitative representation of individual abilities is achieved by establishing an activity indicator system, an award application indicator system, and an individual ability indicator system based on activity data and feature data. This forms a multi-dimensional evaluation system covering students' individual abilities, activity participation, and community management, enabling comprehensive quantitative assessment. A student individual prediction model using SVR+CNN-LSTM+Attention is constructed to predict activity participation probability, achievement of growth records, and credit completion, and to provide early warnings of abnormal behavior. Simultaneously, a community prediction model is built to predict community activity index, quality level, and development trend, enabling intelligent management at the organizational level. The community activity index is automatically calculated daily, and activity frequency limits and downgrade mechanisms are implemented based on departmental and university-level thresholds. It also supports community registration, member assessment, leader election, financial management, activity plan execution monitoring, file management, and reward and punishment mechanisms, achieving standardized and streamlined organizational management. The output module provides second-classroom transcripts, credit and growth record warnings, personalized activity recommendations, activity index reports, and multi-dimensional data visualization. It supports annual weight optimization, cross-university comparison, regional ranking, and indicator anomaly monitoring, achieving visualized and intelligent decision support.

Claims

1. A novel method for predicting and early warning in extracurricular activities, characterized in that: Includes the following steps: Step 1: The data acquisition module collects basic information about universities, structured information about departments and majors, and student registration information through the Internet. Sensitive information is encrypted or masked. Student registration information can be updated and verified annually. Step 2: The behavior data collection module obtains student behavior data and campus card records from the campus system and behavior registration records, summarizes and anonymizes them daily, generates behavior characteristics and time-series characteristics, and calculates the network behavior equilibrium index and anomaly indicators. Step 3: The activity data collection module collects activity data, including information on extracurricular activities, six types of growth records, and activity credit information; Step 4: After cleaning, encoding and dimensionality reduction of the data obtained in Steps 1 to 3, the SHAP method is used to evaluate the importance of features, and finally the feature data is obtained. Step 5: The indicator construction module establishes three major indicator systems based on activity data and feature data, including the activity indicator system, the award application indicator system, and the individual ability indicator system; Step 6: Construct individual student prediction models and tribal prediction models; the individual student prediction model takes feature data as input and outputs activity participation probability, growth record and credit prediction, and abnormal behavior warning; the tribal prediction model takes activity and feature data as input and outputs tribal activity index, quality level and development trend. Step 7: The tribe management module calculates the tribe activity index daily and implements activity limit and downgrade mechanisms based on departmental and university-level thresholds; it also establishes mechanisms for tribe registration, member assessment, leader election, financial management, activity plan monitoring, file management, and rewards and punishments. Step 8: The output module generates extracurricular transcripts, credit and growth record alerts, personalized activity recommendations, activity index reports, and multi-dimensional data visualization results.

2. The novel second-classroom prediction and early warning optimization method as described in claim 1, characterized in that: When collecting student registration information, the data collection module supports updating registration information when students change majors, take a leave of absence, or resume their studies, and conducts annual verification according to the prescribed cycle.

3. The novel second-classroom prediction and early warning optimization method as described in claim 1, characterized in that: In step 2, the behavior data collection module acquires students' access time for educational, entertainment, and gaming activities, as well as campus card records. The behavior data is summarized daily to generate logs, and then anonymized and de-identified to form behavior features. The network behavior equilibrium index and anomaly indicators are further calculated to generate time-series behavior features that capture long-term trends.

4. The novel second-classroom prediction and early warning optimization method as described in claim 1, characterized in that: When performing step 3, the activity data collection module acquires information on extracurricular activities, six types of growth records, and activity credit information. The extracurricular activity information is collected based on student registration information and activity registration records, including activity number, category, time, organizer information, and activity participation records. The six types of growth records include ideology, academics, social practice, volunteer service, cultural and sports qualities, and innovation and entrepreneurship. The activity credit information includes activity type credits and upper limit rules.

5. The novel second-classroom prediction and early warning optimization method as described in claim 1, characterized in that: When performing step 4, the feature engineering module processes the data obtained in steps 1 to 3, including data cleaning, feature encoding, dimensionality reduction and feature importance evaluation. Data cleaning includes mode imputation for categorical features, median imputation for numerical features, and forward imputation for time series features, and automatically marking outlier data to trigger manual review; Feature encoding includes vectorizing student registration information, constructing activity preference vectors, and encoding behavioral features; dimensionality reduction includes using an autoencoder to reduce the dimensionality of the original features and using principal component analysis to reduce the dimensionality of tribal features; feature importance assessment includes using the SHAP method.

6. The novel second-classroom prediction and early warning optimization method as described in claim 1, characterized in that: In step 6, the student individual prediction model is input with feature data and uses an SVR+CNN-LSTM+Attention structure to output activity participation probability, growth record achievement prediction, credit completion prediction, and abnormal behavior warning.

7. The novel second-classroom prediction and early warning optimization method as described in claim 1, characterized in that: When performing step 8, the output module generates extracurricular transcripts, credit warnings, growth record warnings, and personalized activity recommendations. It also generates activity index reports, activity quality visualizations, and member participation data. In addition, it supports annual weight optimization, cross-school comparisons, regional rankings, indicator anomaly monitoring, and multi-dimensional data visualization output.

Images