Student state analysis method and device, storage medium and electronic equipment

Abstract

The invention discloses a student state analysis method and device, a storage medium and electronic equipment, and relates to the technical field of education, and the method comprises the steps: obtaining a multi-dimensional student data set, carrying out the feature extraction of the multi-dimensional student data set through employing a Spark distributed calculation framework, and obtaining a multi-dimensional feature data set corresponding to the multi-dimensional student data set, the multi-dimensional student data set comprises student enrollment data of students and behavior data of the students in a school; carrying out missing value identification on the multi-dimensional feature data set based on a chain equation to obtain missing data in the multi-dimensional feature data set, and carrying out multiple interpolation on the missing data to obtain a target multi-dimensional student data set; constructing a student data sequence according to the target multi-dimensional student data set, converting the student data sequence into a frequency domain data sequence, and determining frequency domain energy distribution of the frequency domain data sequence; the accuracy of positioning the target student in the abnormal student state can be improved, and the risk of student loss is reduced.

Description

Technical Field

[0001] This application relates to the field of educational technology, and in particular to a student status analysis method, apparatus, storage medium, and electronic device. Background Technology

[0002] Analyzing student status during the admissions process allows for the use of data-driven and artificial intelligence technologies to dynamically identify and evaluate student performance, thereby improving the accuracy, fairness, and personalization of admissions services.

[0003] Currently, existing technologies for analyzing student status during the enrollment process mainly rely on human experience and static data analysis. Specifically, they use rule-based judgment or simple statistical methods to process relevant student data, thereby obtaining student status assessment results for enrollment and after enrollment.

[0004] However, student behavior data during school hours is a highly complex, diverse, and temporal data type. The diversity is reflected in the wide range of data sources and the potential for data loss and abnormal fluctuations. The temporal and complex nature is reflected in the fact that students' classroom performance, homework completion, social activities, and data collected by motion-sensing devices all have temporal and complex characteristics. Therefore, student behavior data during school hours is a type of data with complex and dynamically changing relationships.

[0005] Therefore, if rule-based judgment or simple statistical analysis is used to assess student status, it will be difficult to systematically identify and process multi-source data with missing or abnormal data during the assessment process. This will lead to the inability to accurately capture the correlation information in the temporal and complex student data during the analysis process. Consequently, the student status assessment results will be difficult to accurately identify students' abnormal behaviors and psychological states, resulting in delayed intervention for students' poor conditions and an increased risk of student attrition. Summary of the Invention

[0006] In view of this, this application provides a student status analysis method, device, storage medium, and electronic device. The main purpose is to improve the existing technology that uses rule judgment or simple statistical analysis methods to assess student status. This leads to difficulties in systematically identifying and processing multi-source data with missing or abnormal data during the assessment process. Consequently, it is impossible to accurately capture the correlation information in the temporal and complex student data during the analysis process. As a result, the obtained student status assessment results are difficult to accurately identify students' abnormal behaviors and psychological states, leading to delayed intervention in students' poor conditions and an increased risk of student attrition.

[0007] Firstly, this application provides a student status analysis method, including: A multi-dimensional student data set is obtained, and the Spark distributed computing framework is used to extract features from the multi-dimensional student data set to obtain a multi-dimensional feature data set corresponding to the multi-dimensional student data set. The multi-dimensional student data set includes student enrollment data and student behavior data at school. The missing values ​​of the multidimensional feature data set are identified based on the chain equation to obtain the missing data in the multidimensional feature data set. The missing data is then subjected to multiple imputation to obtain the target multidimensional student data set. A student data sequence is constructed based on the target multi-dimensional student data set, the student data sequence is converted into a frequency domain data sequence, and the frequency domain energy distribution of the frequency domain data sequence is determined. Based on the frequency domain energy distribution, abnormal data sequences are identified from the student data sequences, and student status analysis is performed based on the abnormal data sequences to obtain the target students with abnormal student statuses.

[0008] Optionally, the step of identifying missing values ​​in the multi-dimensional feature data set based on chain equations to obtain missing data in the multi-dimensional feature data set, and performing multiple imputation on the missing data to obtain the target multi-dimensional student data set, includes: Determine the initial placeholder value corresponding to each feature data sample in the multi-dimensional feature data set, and construct the dependency matrix corresponding to the multi-dimensional feature data set based on the initial placeholder value; A regression model is constructed based on the dependency matrix, and the regression model is used to predict the missing values ​​in the multi-dimensional feature data set to obtain the missing data in the multi-dimensional feature data set. The multi-dimensional feature data set is divided into at least one multi-dimensional feature data subset, and the missing data in the at least one multi-dimensional feature data subset is independently and randomly sampled and imputed to obtain the imputation results corresponding to the at least one multi-dimensional feature data subset respectively. The interpolation results corresponding to the at least one subset of multi-dimensional feature data are merged to obtain the target multi-dimensional student data set.

[0009] Optionally, merging the interpolation results corresponding to the at least one subset of multi-dimensional feature data to obtain the target multi-dimensional student data set includes: Determine the imputation variance data of the missing data in the at least one multi-dimensional feature data subset during the independent random sampling imputation process, and evaluate the imputation uncertainty data corresponding to the at least one multi-dimensional feature data subset based on the imputation variance data; Based on the interpolation uncertainty data corresponding to the at least one multi-dimensional feature data subset, a weighted analysis is performed on the interpolation results corresponding to the at least one multi-dimensional feature data subset to obtain the target multi-dimensional student data set.

[0010] Optionally, the step of constructing a student data sequence based on the target multi-dimensional student data set, converting the student data sequence into a frequency domain data sequence, and determining the frequency domain energy distribution of the frequency domain data sequence includes: Determine the timestamp information of the data in the target multidimensional student dataset, and convert the target multidimensional student dataset into the student data sequence based on the timestamp information; The student data sequence is segmented based on a sliding window to obtain at least one data sequence segment. The at least one data sequence segment is then processed by a fast Fourier transform to obtain at least one frequency domain data sequence segment corresponding to the at least one data sequence segment. Determine the energy density data corresponding to the at least one frequency domain data sequence segment, and generate the frequency domain energy distribution of the at least one frequency domain data sequence segment based on the energy density data.

[0011] Optionally, the step of identifying abnormal data sequences from the student data sequences based on the frequency domain energy distribution, and performing student status analysis based on the abnormal data sequences to obtain target students with abnormal student statuses, includes: Identify anomalous frequency domain data sequences with energy densities less than an energy density threshold from the frequency domain energy distribution of the at least one frequency domain data sequence segment; Determine the abnormal window corresponding to the abnormal frequency domain data sequence, segment the abnormal frequency domain data based on the abnormal window, and perform short-time Fourier transform processing on the segmented abnormal data sequence to obtain the time-frequency domain data sequence segment corresponding to the abnormal data sequence segment. The abnormal time point and abnormal frequency range corresponding to the abnormal state are determined based on the time-frequency domain data sequence segment. Based on the abnormal time points and the abnormal frequency range, student status analysis is performed to identify target students with abnormal student status.

[0012] Optionally, the step of obtaining a multi-dimensional student dataset and using the Spark distributed computing framework to extract features from the multi-dimensional student dataset to obtain a multi-dimensional feature dataset corresponding to the multi-dimensional student dataset includes: A multi-dimensional student data set is obtained, and features are extracted from the multi-dimensional student data set using the Spark distributed computing framework to obtain a candidate multi-dimensional feature data set. The candidate multidimensional feature data set is filtered based on the gradient optimization algorithm to obtain the multidimensional feature data set corresponding to the multidimensional student data set.

[0013] Secondly, this application provides a student status analysis device, comprising: The acquisition module is configured to acquire a multi-dimensional student data set, and use the Spark distributed computing framework to extract features from the multi-dimensional student data set to obtain a multi-dimensional feature data set corresponding to the multi-dimensional student data set. The multi-dimensional student data set includes student enrollment data and student behavior data at school. The identification module is configured to identify missing values ​​in the multi-dimensional feature data set based on chain equations, obtain missing data in the multi-dimensional feature data set, and perform multiple imputation on the missing data to obtain the target multi-dimensional student data set. The construction module is configured to construct a student data sequence based on the target multi-dimensional student data set, convert the student data sequence into a frequency domain data sequence, and determine the frequency domain energy distribution of the frequency domain data sequence; The identification module is also configured to identify abnormal data sequences from the student data sequences based on the frequency domain energy distribution, and to perform student status analysis based on the abnormal data sequences to obtain target students with abnormal student statuses.

[0014] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the student status analysis method described in the first aspect.

[0015] Fourthly, this application provides an electronic device, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor executes the computer program to implement the student status analysis method described in the first aspect.

[0016] By employing the above technical solutions, this application provides a student status analysis method, device, storage medium, and electronic device. Compared with existing technologies, this application achieves multi-dimensional feature mining of student enrollment data and on-campus behavior data by acquiring multi-dimensional student datasets and using the Spark distributed computing framework for feature extraction, thus adapting to the needs of data diversity. It improves the integrity of the target multi-dimensional student dataset by identifying missing values ​​and performing multiple imputations based on chain equations, solving the problem of missing data from multiple sources. By constructing student data sequences based on the target multi-dimensional student dataset and converting them into frequency domain data sequences, and determining the frequency domain energy distribution, it effectively transforms the temporal characteristics of student data into frequency domain features, accurately capturing the correlation information in temporal data. Furthermore, by identifying abnormal data sequences based on the frequency domain energy distribution and performing student status analysis, it improves the accuracy of identifying abnormal student behavior and psychological states in highly complex, diverse, and temporally integrated student data, thereby reducing the risk of student attrition. Attached Figure Description

[0017] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A flowchart illustrating a student status analysis method provided in an embodiment of this application is shown; Figure 2 A flowchart illustrating a student status analysis method provided in an embodiment of this application is shown; Figure 3 This illustration shows a schematic diagram of an algorithm training application environment for Spark that combines chain equations and fast Fourier transform, as provided in an embodiment of this application. Figure 4 This illustration shows a flowchart of an algorithm training method combining chain equations and fast Fourier transform provided in an embodiment of this application. Figure 5 This paper shows a schematic diagram of the structure of a student status analysis device provided in an embodiment of this application; Figure 6 A schematic diagram of the structure of an electronic device provided in an embodiment of this application is shown. Detailed Implementation

[0020] The embodiments of this application will now be described in more detail with reference to the accompanying drawings. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0021] To address the shortcomings of existing technologies for student status analysis, which struggle to systematically identify and process multi-source data, and the highly complex and time-series nature of student behavior data during school hours (such as classroom performance, homework completion, social activities, and sensory device data), this embodiment provides a student status analysis method. However, relying solely on rule-based judgments or simple statistics fails to accurately identify abnormal behaviors and psychological states, leading to delayed interventions and an increased risk of student attrition. Figure 1 As shown, the method includes: Step 101: Obtain a multi-dimensional student dataset and use the Spark distributed computing framework to extract features from the multi-dimensional student dataset to obtain a multi-dimensional feature dataset corresponding to the multi-dimensional student dataset.

[0022] The multi-dimensional student data set includes student enrollment data and student behavior data at school.

[0023] In this embodiment, the multi-dimensional student data set can be a collection of data related to students in various aspects during the enrollment process and their time at school. The data sources for the multi-dimensional student data set can be the result of integrating multiple data collection channels such as the school's enrollment system, teaching management system, and campus life monitoring system. The data collected from multiple data collection channels has high complexity, diversity, and time sequence. For example, the multi-dimensional student data set in this embodiment can specifically include students' enrollment data and students' behavior data at school. Enrollment data can include, but is not limited to, data directly related to the enrollment process, such as students' enrollment contact records, exam scores, payment information, and complaint feedback. Students' behavior data at school can include, but is not limited to, classroom performance data (such as classroom interaction frequency, concentration-related data, etc.), homework trajectory data (such as homework submission time, completion quality, and types of incorrect questions, etc.), social activity data (such as frequency of peer interaction on campus, participation in club activities, etc.), data collected by motion sensing devices (such as exercise duration, physical condition-related monitoring data, etc.), and can also include various training and examination data such as knowledge competitions, sports, and arts in primary, secondary, and high schools.

[0024] In this embodiment, the Spark distributed computing framework can be a distributed computing framework for large-scale data processing. The Spark distributed computing framework has the ability to efficiently process massive amounts of data, supports parallel computing and distributed storage interaction, and can adapt to massive data scenarios that may exist in multi-dimensional student datasets. For example, in this embodiment, the relevant functional modules of the Spark distributed computing framework can be used to read multi-dimensional student datasets and perform distributed data processing, thereby improving data processing efficiency.

[0025] In this embodiment, the multi-dimensional feature data set can be a collection of feature data extracted from a multi-dimensional student data set, reflecting different dimensions of student attributes and states. This feature data can provide a foundation for subsequent missing value processing, anomaly identification, and student status analysis. For example, the multi-dimensional feature data extracted in this embodiment may include student performance characteristics, special talent characteristics, behavioral habit characteristics, and interaction characteristics, such as exam score distribution characteristics, sports or arts talent level characteristics, homework completion pattern characteristics, and frequency of admissions communication characteristics.

[0026] In this embodiment of the application, a multi-dimensional student dataset is obtained, and features are extracted from the multi-dimensional student dataset using the Spark distributed computing framework to obtain the multi-dimensional feature dataset corresponding to the multi-dimensional student dataset. First, the multi-dimensional student dataset to be processed and the feature subset to be used for algorithm training are determined, and corresponding computing resources are configured. The original dataset serves as the input dataset for feature engineering, and subsequent training and testing sets can be partitioned for model training and validation, respectively. Then, the multi-dimensional student dataset can be read using the Spark distributed computing framework to obtain the various feature data types of the dataset. Finally, features can be extracted from the multi-dimensional student dataset using Spark-related algorithms to obtain a candidate multi-dimensional feature dataset.

[0027] Step 102: Identify missing values ​​in the multi-dimensional feature data set based on the chain equation to obtain the missing data in the multi-dimensional feature data set, and perform multiple imputation on the missing data to obtain the target multi-dimensional student data set.

[0028] In the embodiments of this application, the chain equation can be a statistical equation used to handle missing values ​​in a dataset. The chain equation can achieve accurate identification and reasonable imputation of missing data by constructing dependencies between variables and iteratively predicting missing values ​​using existing data. For example, the chain equation can fully utilize the correlations between multi-dimensional feature data, avoiding bias caused by imputation of a single variable, and is suitable for complex missing data situations that may exist in multi-dimensional feature data sets.

[0029] In the embodiments of this application, missing value identification can be achieved by using a specific algorithm or method to detect whether there are missing data in a multi-dimensional feature data set, and to locate the specific location of the missing data and its corresponding feature dimension. For example, in the embodiments of this application, missing value identification may specifically involve discovering situations such as some students' exam scores not being recorded, or missing homework trajectory data for a certain period of time.

[0030] In this embodiment of the application, missing data may be a portion of data that was not collected or was not fully recorded in the multi-dimensional feature data set. Missing data may affect the accuracy and completeness of subsequent analysis results. For example, missing data in this embodiment may specifically include missing records of students' artistic talent training, or the failure to retain some feedback information from admissions communication.

[0031] In the embodiments of this application, multiple imputation can be a process of generating multiple complete datasets through multiple independent imputation processes for missing data, and then merging the analysis results of these datasets. For example, in the embodiments of this application, multiple imputation can specifically fill in missing values ​​through multiple random samplings, fully considering the randomness and diversity of data distribution.

[0032] In this embodiment, the target multi-dimensional student dataset can be a dataset whose data integrity is guaranteed after missing value identification and multiple imputation processing. This target multi-dimensional student dataset can provide a data foundation for subsequent anomaly detection and student status analysis. For example, in this embodiment, every feature data item for each student in the target multi-dimensional student dataset can be fully supplemented, and there are no key missing items that would affect the analysis.

[0033] In this embodiment, missing values ​​are identified in a multi-dimensional feature data set based on a chain equation to obtain missing data in the multi-dimensional feature data set. Multiple imputation is then performed on the missing data to obtain the target multi-dimensional student data set. This can be achieved by first identifying the missing values ​​in the multi-dimensional feature data set through the relevant logic of the chain equation, accurately locating the position and related feature information of the missing data; then, multiple imputation can be used to fill in the missing data. Through multiple independent imputation and result merging, a complete and reliable target multi-dimensional student data set is obtained.

[0034] Step 103: Construct a student data sequence based on the target multi-dimensional student data set, convert the student data sequence into a frequency domain data sequence, and determine the frequency domain energy distribution of the frequency domain data sequence.

[0035] In this embodiment, the student data sequence can be an ordered data sequence formed by sorting a target multi-dimensional student data set according to specific rules (such as chronological order). The student data sequence can be used to reflect the changing patterns of student data over time or other dimensions to adapt to the temporal characteristics of student data. For example, the student data sequence in this embodiment may include daily homework completion data and weekly admissions communication records arranged in timestamp order.

[0036] In this embodiment, the frequency domain data sequence can be a data sequence obtained by transforming a time-domain student data sequence into the frequency domain using a specific transformation algorithm. The frequency domain data sequence can be used to reflect the changing characteristics of student data from a frequency perspective, addressing the complexity of student data and identifying abnormal fluctuations or hidden patterns in the data. For example, the frequency characteristics corresponding to periodic anomalies or sudden fluctuations in student behavior data can be discovered through the frequency domain data sequence.

[0037] In the embodiments of this application, the frequency domain energy distribution can be the distribution of energy magnitude corresponding to each frequency component in the frequency domain data sequence. The energy level of different frequency components can reflect the proportion and influence of the corresponding frequency features in the original data. For example, if the energy in a frequency range in the frequency domain energy distribution is too high or too low, it may correspond to an abnormal change pattern in the original student data.

[0038] In this embodiment, a student data sequence is constructed based on a target multi-dimensional student dataset. The student data sequence is then converted into a frequency domain data sequence, and the frequency domain energy distribution of the frequency domain data sequence is determined. This can be achieved by first converting the target multi-dimensional student dataset into an ordered student data sequence based on the timestamp information or other preset rules of the data in the target multi-dimensional student dataset. For non-numerical data, digitization encoding can be performed first to eliminate the influence of dimensions. Then, the student data sequence is segmented based on a preset sliding window to obtain at least one data sequence segment. Next, a Fast Fourier Transform (FFT) is applied to each data sequence segment to convert the time-domain data sequence segment into a frequency-domain data sequence segment. Finally, the energy density data corresponding to each frequency-domain data sequence segment is calculated, and the frequency domain energy distribution of the frequency-domain data sequence segment is generated based on these energy density data.

[0039] Step 104: Identify abnormal data sequences from the student data sequences based on the frequency domain energy distribution, and perform student status analysis based on the abnormal data sequences to obtain the target students with abnormal student statuses.

[0040] In this embodiment of the application, the abnormal data sequence can be a part of the data sequence in which the frequency domain energy distribution does not conform to the normal data pattern. The abnormal data sequence can be used to reflect the abnormal behavior or state changes of students in a specific period.

[0041] In this embodiment of the application, student status analysis can be a process of comprehensively analyzing and judging students' psychological state, learning attitude, and willingness to enroll based on identified abnormal data sequences and students' multi-dimensional characteristic data. For example, the student status analysis in this embodiment of the application may specifically include judging whether a student has a tendency to dislike school, potential mental health issues, thoughts of dropping out, or a decreased willingness to enroll.

[0042] In this embodiment, an abnormal student state can refer to a student's deviation from the normal state during learning, life, or enrollment coordination. An abnormal student state may affect a student's academic development or enrollment decisions, and timely intervention and guidance can be provided for students in abnormal states. For example, an abnormal student state in this embodiment may specifically include aversion to school, anxiety, decreased school acceptance, or a tendency to drop out.

[0043] In this embodiment of the application, the target student can be an individual student whose abnormal student status is determined after student status analysis. Targeted intervention for the target student can reduce the risk of student attrition.

[0044] In this embodiment, abnormal data sequences are identified from student data sequences based on frequency domain energy distribution. Student status analysis is then performed based on these abnormal data sequences to identify target students with abnormal student states. First, based on a preset energy density threshold, abnormal frequency domain data sequences with energy densities exceeding the normal range (e.g., less than the energy density threshold) are identified from the frequency domain energy distribution of each frequency domain data sequence segment. Then, the abnormal windows corresponding to these abnormal frequency domain data sequences are determined, and the original student data sequence is segmented based on these abnormal windows to obtain abnormal data sequence segments. Next, short-time Fourier transform (STFT) is applied to the segmented abnormal data sequences for time-frequency joint analysis to accurately capture abnormal times and features in complex time-series data, resulting in time-frequency domain data sequence segments corresponding to the abnormal data sequences. Then, based on the time-frequency domain data sequence segments, the abnormal time points and abnormal frequency ranges corresponding to the abnormal states are accurately determined. Finally, a comprehensive student status analysis is performed by combining the abnormal time points, abnormal frequency ranges, and multi-dimensional characteristic data of the students. For student data with complexity, diversity, and temporal sequence characteristics, it is determined whether students exhibit aversion to learning, thoughts of dropping out, or psychological abnormalities, thus identifying target students with abnormal student states.

[0045] Compared with existing technologies, this embodiment acquires multi-dimensional student datasets and uses the Spark distributed computing framework for feature extraction, enabling multi-dimensional feature mining of student enrollment data and on-campus behavior data, thus adapting to the needs of data diversity. By identifying missing values ​​and performing multiple imputations on the multi-dimensional feature data set based on chain equations, the integrity of the target multi-dimensional student dataset is improved, solving the problem of missing data from multiple sources. By constructing student data sequences based on the target multi-dimensional student dataset and converting them into frequency domain data sequences, and determining the frequency domain energy distribution, the effective transformation of temporal features of student data into frequency domain features is achieved, accurately capturing the correlation information in temporal data. By identifying abnormal data sequences based on the frequency domain energy distribution and performing student status analysis, the accuracy of identifying abnormal student behavior and psychological states in highly complex, diverse, and temporally integrated student data can be improved, thereby reducing the risk of student attrition.

[0046] As an optional approach, when performing the task of "identifying missing values ​​in a multi-dimensional feature dataset based on chain equations, obtaining missing data in the multi-dimensional feature dataset, and performing multiple imputation on the missing data to obtain the target multi-dimensional student dataset," the following methods can be used, but are not limited to these: Figure 2 As shown, it includes: Step 201: Determine the initial placeholder value for each feature data sample in the multi-dimensional feature data set, and construct the dependency matrix corresponding to the multi-dimensional feature data set based on the initial placeholder value.

[0047] In this embodiment, the initial placeholder value can be a temporary replacement value assigned to a sample of missing data in a multi-dimensional feature data set. The initial placeholder value can be used to fill in the missing positions. For example, the initial placeholder value in this embodiment may include the mean, median, or other reasonable values ​​determined based on the data distribution pattern of the corresponding feature data.

[0048] In this embodiment, the dependency matrix can be a matrix used to characterize the dependencies between various variables (features) in a multi-dimensional feature dataset. The dependency matrix can capture the complex associations between variables caused by the complexity of student data. The rows of the dependency matrix can represent variables to be imputed, and the columns can represent predictor variables. The dependency matrix can be used to present the mutual influence relationships between variables. For example, in this embodiment, the dependency matrix can use numerical values ​​or specific identifiers to represent the strength of the dependency between variables, providing a basis for constructing a regression model.

[0049] In this embodiment, an initial placeholder value is determined for each feature data sample in the multi-dimensional feature data set. Based on the initial placeholder value, a dependency matrix corresponding to the multi-dimensional feature data set is constructed. First, an initial placeholder value is assigned to the possible missing values ​​of each feature data sample in the multi-dimensional feature data set, such as calculating the mean or median of each feature data as the initial placeholder value for the corresponding missing position. Then, based on the inherent correlation between each feature variable in the multi-dimensional feature data set, a dependency matrix X is constructed to clarify the correspondence between the variable to be imputed and the predicted variable.

[0050] Step 202: Construct a regression model based on the dependency matrix, and use the regression model to predict the missing values ​​in the multi-dimensional feature data set to obtain the missing data in the multi-dimensional feature data set.

[0051] In this embodiment, the regression model can be a mathematical model built based on a dependency matrix for predicting missing values. The regression model can fully utilize the complex relationships between variables in student data through iterative prediction, avoiding bias caused by a single imputation method and adapting to the needs of data diversity. The regression model can be used to calculate the missing values ​​of the variable to be imputed using known data of the predicted variable, reflecting the linear or non-linear relationship between variables. For example, the regression model in this embodiment can be a linear regression model or a non-linear regression model built based on machine learning algorithms, and the choice can be made according to the data characteristics and dependency type.

[0052] For the embodiments of this application, the regression model can be as shown in Formula 1, where yi can represent the predicted missing value. It can represent the intercept term. When all independent variables Xk=0, the theoretical value of yi can reflect the baseline level of the model. It can represent linear combinations. It can represent the regression coefficient of the k-th variable. It can represent the expected change in yi when Xk changes by 1 unit (with other variables remaining unchanged), and Xk can represent the k-th independent variable value of the i-th observation (such as characteristic data such as student name, age, etc.). It can represent the random error term. This can represent random disturbances not captured by the model (such as the effects of unobserved quantities), and can be assumed to be... .

[0053] (Formula 1) In this embodiment of the application, a regression model is constructed based on the dependency matrix. The regression model is used to predict the missing values ​​in the multi-dimensional feature data set. The missing data in the multi-dimensional feature data set can be obtained by selecting an appropriate regression model type based on the dependency matrix and constructing it. Then, the known data of the predictor variable can be input into the regression model to predict the missing values ​​in the multi-dimensional feature data set and obtain the predicted values ​​at each missing position. The predicted values ​​can then replace the previously set initial placeholder values, and the above prediction and replacement process can be repeated until the model converges (convergence is usually achieved in 5-10 iterations) to obtain the prediction results of the missing data in the multi-dimensional feature data set.

[0054] Step 203: Divide the multi-dimensional feature data set into at least one multi-dimensional feature data subset, and independently and randomly sample and impute the missing data in the at least one multi-dimensional feature data subset to obtain the imputation results corresponding to the at least one multi-dimensional feature data subset.

[0055] In this embodiment, the multi-dimensional feature data subset can be several subsets obtained by dividing the complete multi-dimensional feature data set according to a preset partitioning rule (such as random partitioning, partitioning by feature dimension, etc.). Each multi-dimensional feature data subset can retain some features and sample information of the original data set. For example, the original data set can be randomly divided into m independent multi-dimensional feature data subsets, and the sample distribution of each subset is roughly the same as that of the original data set.

[0056] In this embodiment, independent random sampling imputation can be performed by using an independent random sampling method to extract values ​​from a reasonable distribution range of the data for missing data in each subset of multi-dimensional feature data. For example, independent random sampling imputation can be based on the probability distribution of the corresponding feature data, so that the imputed values ​​conform to the natural distribution pattern of the data.

[0057] In this embodiment, the multi-dimensional feature data set is divided into at least one multi-dimensional feature data subset. Missing data in each of the at least one multi-dimensional feature data subset is imputed by independent random sampling. This process can be achieved by first dividing the multi-dimensional feature data set into at least one multi-dimensional feature data subset based on the actual data volume and processing requirements, such as dividing it into m complete datasets D1, ..., Dm. Then, for each multi-dimensional feature data subset, missing data can be imputed using independent random sampling. The imputation process can be based on the data distribution characteristics and dependencies. After imputing the missing data in each subset, the imputation result corresponding to that subset can be obtained, meaning each subset becomes a complete dataset.

[0058] Step 204: Merge the interpolation results corresponding to at least one subset of multi-dimensional feature data to obtain the target multi-dimensional student data set.

[0059] For the embodiments of this application, the Rubin rule can be used to merge the interpolation results corresponding to at least one subset of multi-dimensional feature data. The specific calculation formulas are shown in Formulas 2 and 3, where... It can represent a sample estimate, that is, an estimate of population parameters (such as mean, variance, etc.), and m can represent the number of multidimensional feature data subsets, that is, the number of independent experiments or observations (such as the number of repeated samplings, the number of Monte Carlo simulations). It can represent the estimated value corresponding to the interpolation result of the i-th multi-dimensional feature data subset (such as the statistic of a single sampling). It can represent the standard error of the estimator. Can be used for measurement The volatility reflects the accuracy of the estimation. The smaller the value, the more stable and accurate it is. It can represent the standard error corresponding to the interpolation result of the i-th multidimensional feature data subset (such as the standard error of the statistic in a single sampling).

[0060] (Formula 2) (Formula 3) In the embodiments of this application, merging the interpolation results corresponding to at least one multi-dimensional feature data subset to obtain the target multi-dimensional student data set can be achieved by first performing separate statistical analysis on the interpolation results of each multi-dimensional feature data subset and calculating the estimated value corresponding to each subset. and standard error Then, the combined estimate of all subset interpolation results can be calculated according to the Rubin rule formula. and combined standard error Based on the merged estimation results, a complete target multidimensional student data set can be generated.

[0061] As an optional approach, when performing the task of "merging the imputation results corresponding to at least one subset of multi-dimensional feature data to obtain a target multi-dimensional student dataset", the following methods can be used, but are not limited to: determining the imputation variance data of the missing data in at least one subset of multi-dimensional feature data during the independent random sampling imputation process; evaluating the imputation uncertainty data corresponding to at least one subset of multi-dimensional feature data based on the imputation variance data; and performing a weighted analysis on the imputation results corresponding to at least one subset of multi-dimensional feature data based on the imputation uncertainty data corresponding to at least one subset of multi-dimensional feature data to obtain the target multi-dimensional student dataset.

[0062] In this embodiment, the imputation variance data can be the variance between multiple imputed values ​​obtained after independently and randomly sampling and imputing the same missing data from multiple subsets of multi-dimensional feature data. The imputation variance data reflects the dispersion of the imputation results and can quantify the imputation complexity caused by the diversity of student data by adapting to uncertainty. A larger imputation variance data indicates a greater difference in the imputation results of different subsets, and a higher level of imputation uncertainty.

[0063] In this embodiment, the interpolation uncertainty data can be quantitative data used to characterize the reliability of the interpolation result, obtained based on the interpolation variance data. The interpolation uncertainty data can reflect the uncertainty of the result caused by data randomness and model error during the interpolation process. For example, the interpolation uncertainty data in this embodiment can be calculated by combining the interpolation variance data with data distribution characteristics; a larger value indicates lower reliability of the interpolation result.

[0064] In this embodiment, weighted analysis can assign different weights to the imputation results of each subset based on the imputation uncertainty data corresponding to each multi-dimensional feature data subset. The subset with lower imputation uncertainty has a larger weight and a higher proportion in the weighted analysis merging result. For example, weighted analysis can use a weighted average method to calculate the imputation results of each subset according to the corresponding weights to obtain the final imputation value.

[0065] In this embodiment, the imputation values ​​of all multi-dimensional feature data subsets for each missing data location can be collected first, and the imputation variance data of the imputation values ​​can be calculated. Then, based on the imputation variance data, combined with the overall distribution pattern of the data and the characteristics of the model, the imputation uncertainty data corresponding to each multi-dimensional feature data subset can be evaluated. After that, weights can be assigned to each subset according to the imputation uncertainty data, with the subsets with lower imputation uncertainty being assigned greater weights. Finally, the imputation results of all multi-dimensional feature data subsets can be weighted according to the assigned weights to obtain the final imputation value for each missing data location, thereby integrating them to form a complete target multi-dimensional student data set.

[0066] As an optional approach, when performing the task of "constructing a student data sequence based on a target multi-dimensional student dataset, converting the student data sequence into a frequency domain data sequence, and determining the frequency domain energy distribution of the frequency domain data sequence," the following methods can be used, but are not limited to: determining the timestamp information of the data in the target multi-dimensional student dataset, converting the target multi-dimensional student dataset into a student data sequence based on the timestamp information; segmenting the student data sequence based on a sliding window to obtain at least one data sequence segment, performing a fast Fourier transform on the at least one data sequence segment to obtain at least one frequency domain data sequence segment corresponding to the at least one data sequence segment; determining the energy density data corresponding to the at least one frequency domain data sequence segment, and generating the frequency domain energy distribution of the at least one frequency domain data sequence segment based on the energy density data.

[0067] In this embodiment, the timestamp information can be the identification information of the collection time or occurrence time corresponding to each data record in the target multi-dimensional student dataset. The timestamp information can be used to reflect the time attribute of the data and provide a sorting basis for the construction of the data sequence. For example, the timestamp information in this embodiment may specifically include date and time format, or a sequence number assigned according to time order.

[0068] In the embodiments of this application, a sliding window can be a data segmentation tool. A sliding window can be used to segment a continuous sequence of student data according to a preset window size and sliding step size, dividing long sequence data into multiple continuous short data segments. The sliding window can preserve the temporal continuity of student data and adapt to the temporal characteristics of data. The size of the sliding window can be set according to the sampling theorem.

[0069] In the embodiments of this application, the Fast Fourier Transform (FFT) can be a highly efficient Fourier Transform algorithm. The Fast Fourier Transform can be used to quickly convert discrete signals in the time domain into frequency domain signals, reducing computational complexity and making it suitable for frequency domain conversion processing of massive student data sequences. For example, the Fast Fourier Transform algorithm in the embodiments of this application can convert the time dimension features in the student data sequence into frequency dimension features, facilitating the extraction of abnormal information from a frequency perspective.

[0070] In the embodiments of this application, the energy density data can be the energy density value corresponding to each frequency component in the frequency domain data sequence. The energy density data can be used to quantify the energy intensity of each frequency component. For example, in the embodiments of this application, the energy density data can be obtained by calculating the square of the amplitude of each component in the frequency domain data sequence, which can reflect the energy strength of the corresponding frequency characteristics.

[0071] In this embodiment, timestamp information corresponding to each data point can be extracted from the target multi-dimensional student data set. All data are sorted according to the order of the timestamps, and the unordered target multi-dimensional student data set is transformed into an ordered student data sequence. For non-numerical data (such as student feedback evaluations, behavioral tags, etc.), digital encoding processing can be performed first to convert them into a computable numerical form. Then, an appropriate sliding window size and sliding step can be set according to the sampling theorem. Based on the sliding window, the ordered student data sequence is segmented, dividing the long sequence into multiple continuous and potentially overlapping local data sequence segments, ensuring that each data sequence segment can completely retain the local temporal characteristics of the original data.

[0072] In this embodiment, the student data sequence is segmented based on a sliding window to obtain at least one data sequence segment. A Fast Fourier Transform (FFT) is then performed on the at least one data sequence segment to obtain at least one frequency domain data sequence segment corresponding to the at least one data sequence segment. The calculation formula for the FFT is shown in Formula 4, where X[k] represents the k-th component of the frequency domain output sequence, i.e., the complex amplitude of the signal at frequency k / N, and x[n] represents the n-th sample of the time domain input sequence, i.e., a finite discrete signal. The complex exponential spiral factor can be represented by the imaginary unit j, which can represent the orthogonal frequency domain components. It can correspond to discrete frequencies k / N, and the negative sign in the complex exponential spiral factor can represent the forward transform (positive in the inverse transform IDFT); Formula 4 can be used to convert each time domain data sequence segment into the corresponding frequency domain data sequence segment.

[0073] (Formula 4) For the embodiments of this application, the energy density data of each frequency component in each frequency domain data sequence segment can be calculated. The calculation formula for the energy density data can be as shown in Formula 5, where E[k] can represent the energy density of the k-th frequency component. Then, based on the energy density data of all frequency components, the frequency domain energy distribution corresponding to each frequency domain data sequence segment is constructed.

[0074] (Formula 5) As an optional approach, when performing the task of "identifying abnormal data sequences from student data sequences based on frequency domain energy distribution, and performing student status analysis based on the abnormal data sequences to obtain target students with abnormal student status," the following methods can be used, but are not limited to: identifying abnormal frequency domain data sequences with energy densities less than an energy density threshold from the frequency domain energy distribution of at least one frequency domain data sequence segment; determining the abnormal window corresponding to the abnormal frequency domain data sequence, segmenting the abnormal frequency domain data based on the abnormal window, performing short-time Fourier transform processing on the segmented abnormal data sequence to obtain the time-frequency domain data sequence segment corresponding to the abnormal data sequence segment; determining the abnormal time point and abnormal frequency range corresponding to the abnormal status based on the time-frequency domain data sequence segment; and performing student status analysis based on the abnormal time point and abnormal frequency range to obtain the target students with abnormal student status.

[0075] In this embodiment, the Short-Time Fourier Transform (STFT) can be a time-frequency analysis algorithm. The STFT performs a Fourier transform on the signal within a sliding window, simultaneously reflecting both the time and frequency characteristics of the signal, facilitating precise location of the time point and corresponding frequency range of anomaly signals. For example, the STFT in this embodiment can compensate for the shortcomings of the Fast Fourier Transform in time localization, enabling joint time-frequency analysis of anomalous data sequences.

[0076] In this embodiment, the energy density threshold can be a threshold determined based on the statistical distribution of frequency domain energy in normal data. The energy density threshold can be used to determine whether the energy density in a frequency domain data sequence is within the normal range. The energy density threshold is set and adjusted based on expert experience or statistical data results. For example, in this embodiment, the energy density threshold can be specifically set to the 95th percentile of the normal data frequency domain energy distribution. When the energy density of a frequency component exceeds or falls below the energy density threshold, the data sequence corresponding to that frequency component can be determined to be abnormal.

[0077] In this embodiment, the abnormal window can be the original data window in the student data sequence corresponding to the abnormal frequency domain data sequence, that is, the window position and range corresponding to the local data sequence segment of the abnormal frequency domain data sequence generated after the sliding window segmentation. For example, the abnormal window in this embodiment can be derived by reverse derivation from the sequence number of the abnormal frequency domain data sequence and the setting parameters of the sliding window, and the abnormal window can locate the position of the abnormal data segment in the original student data sequence.

[0078] In the embodiments of this application, the abnormal time point can be the specific time point at which the abnormal state begins to occur or occurs in a concentrated manner in the abnormal data sequence.

[0079] In this embodiment, the abnormal frequency range can be the frequency interval corresponding to the abnormal data sequence in the frequency domain. The energy density of the frequency components within the abnormal frequency range may not conform to normal patterns, and the abnormal frequency range can be used to reflect the frequency characteristics of the abnormal state. For example, in this embodiment, the abnormal frequency range can specifically be a high-frequency interval corresponding to a student's short-term sudden abnormal behavior, or a low-frequency interval corresponding to a long-term abnormal trend.

[0080] In this embodiment, the frequency domain energy distribution of all normal data can be statistically analyzed first, and an energy density threshold can be determined based on the statistical results, such as selecting the 95th percentile of the frequency domain energy distribution of normal data as the energy density threshold. Then, the frequency domain energy distribution of each frequency domain data sequence segment can be analyzed, and frequency domain data sequences with energy densities less than the energy density threshold can be identified as abnormal frequency domain data sequences. Based on the relevant information of the abnormal frequency domain data sequence (such as sequence number, corresponding sliding window parameters, etc.), the abnormal window corresponding to the abnormal frequency domain data sequence in the original student data sequence can be determined. Based on the range of the abnormal window, the original student data sequence can be segmented to obtain the abnormal data sequence segment corresponding to the abnormal frequency domain data sequence.

[0081] In the embodiments of this application, a Short-Time Fourier Transform (STFT) is applied to the truncated abnormal data sequence segment. The calculation formula for the Short-Time Fourier Transform is shown in Formula 6, where, It can represent the time-frequency domain output function, and can represent the signal shifted in time. and angular frequency The complex amplitude distribution under the given conditions; x(t) can represent the original input signal, that is, a continuous-time function; It can represent a time-shifting window function. It can be used for localization analysis, and g(t) can represent an odd function (such as a Gaussian window). It can represent controlling time shift; It can represent the complex exponential twitch factor, and j can represent the orthogonal frequency domain component. It can represent angular frequency, and the negative sign can correspond to the positive transform (positive in the inverse transform); by performing time-frequency analysis on the abnormal data sequence segment using Formula 6, the time-frequency domain data sequence segment corresponding to the abnormal data sequence segment can be obtained. The time-frequency domain data sequence segment can reflect the time and frequency characteristics of the abnormal data.

[0082] (Formula 6) In this embodiment of the application, determining the abnormal time point and abnormal frequency range corresponding to the abnormal state based on the time-frequency domain data sequence segment can involve extracting key information from the time-frequency domain data sequence segment to determine the specific abnormal time point of the abnormal state (such as the start time of the abnormal behavior, the duration of the abnormal state, etc.) and the corresponding abnormal frequency range (such as the frequency interval corresponding to the abnormal feature). A comprehensive student status analysis is then performed by combining the abnormal time point, abnormal frequency range, and other multi-dimensional feature data (such as student academic data, social data, and enrollment communication data) from the target multi-dimensional student data set to determine whether students exhibit abnormal states such as aversion to learning, mental illness, or thoughts of dropping out, thereby identifying target students with abnormal student states. It should be noted that manual intervention can also be incorporated into the anomaly judgment and student status analysis process, based on the experience of business experts, to avoid misjudging sparse normal data as abnormal data and ensure the accuracy of target student identification.

[0083] As an optional approach, when performing the task of "obtaining a multi-dimensional student dataset, using the Spark distributed computing framework to extract features from the multi-dimensional student dataset, and obtaining a multi-dimensional feature dataset corresponding to the multi-dimensional student dataset", the following methods can be used, but are not limited to: obtaining a multi-dimensional student dataset, using the Spark distributed computing framework to extract features from the multi-dimensional student dataset, and obtaining a candidate multi-dimensional feature dataset; and using a gradient optimization algorithm to filter features from the candidate multi-dimensional feature dataset, and obtaining a multi-dimensional feature dataset corresponding to the multi-dimensional student dataset.

[0084] In the embodiments of this application, the candidate multi-dimensional feature data set can be a feature data set obtained after preliminary feature extraction of the multi-dimensional student data set through the Spark distributed computing framework. The candidate multi-dimensional feature data set may contain some redundant features that have a low contribution to the student status analysis, and the candidate multi-dimensional feature data set can be further screened and optimized.

[0085] In the embodiments of this application, the gradient optimization algorithm can be an optimization algorithm based on the idea of ​​gradient descent. The gradient optimization algorithm can continuously adjust parameters or screening criteria by calculating the gradient direction of the objective function to achieve the minimization of the objective function or the optimal selection of feature importance. The gradient optimization algorithm can be used for feature screening of candidate multi-dimensional feature datasets. It can filter out core features that contribute highly to anomaly identification from massive and complex features, adapting to the feature redundancy problem caused by the complexity of student data. For example, the gradient optimization algorithm in the embodiments of this application may include, but is not limited to, the AdaGrad algorithm, the SGD algorithm, and the Adam algorithm.

[0086] In this embodiment, a multi-dimensional student dataset covering student enrollment data and on-campus behavior data can be obtained through data collection channels and imported into the Spark distributed computing framework. The parallel computing capabilities of the Spark framework are then used to extract features from the multi-dimensional student dataset. Feature extraction can include basic feature extraction and derived feature construction, resulting in a candidate multi-dimensional feature dataset containing a large number of features. A suitable gradient optimization algorithm (such as the AdaGrad algorithm) can then be selected, and a target function for feature selection can be set (e.g., the contribution of a feature to student status analysis). The importance or gradient information of each feature in the candidate multi-dimensional feature dataset is calculated using the gradient optimization algorithm. Features with high contribution to student status analysis and low redundancy can be selected based on the importance or gradient information of each feature. Finally, the selected features can be combined to form a multi-dimensional feature dataset, providing a high-quality feature foundation for subsequent missing value processing and outlier identification.

[0087] It should be noted that during feature extraction and filtering, the optimal Spark Shuffle mechanism can be set according to data distribution, resource load, and algorithm rules to maximize the use of machine resources and improve the efficiency of feature processing. During training, different parameter combinations can be set, and parameter samples can be iterated. Combined with subsequent missing value handling and outlier detection algorithms, the algorithm can achieve rapid convergence. When exporting the algorithm model after training, the Population Stability Index (PSI) can be used to monitor the stability of the model prediction. When a certain threshold is exceeded, the algorithm can be intervened or retrained in a timely manner. At the same time, all sample sets are iterated, and the optimal parameter configuration combination is obtained by adjusting the relevant hyperparameters.

[0088] Optionally, embodiments of this application also provide a diagram illustrating an algorithm training application environment for Spark that combines chain equations and Fast Fourier Transform, such as... Figure 3 As shown, Figure 3Specifically, it can include features 1 to features n (that is, specific features in the multi-dimensional feature data set in the embodiments of this application), each feature can correspond to multi-threaded concurrent computation processing; multi-threaded concurrent computation processing can correspond to multiple computing nodes in the Spark cluster (including computing nodes 1, 2, etc. up to node n), and the driving nodes in the cluster can interact with the computing nodes to realize distributed concurrent computation of multiple features.

[0089] Optionally, embodiments of this application also provide a flowchart of the training algorithm for Spark combining chain equations and fast Fourier transform, such as... Figure 4 As shown, Figure 4 The specific steps may include: Step 1: Determine the dataset to be processed (i.e., the multi-dimensional student data set in this embodiment), the feature subset to be used for algorithm training (i.e., the candidate multi-dimensional feature data set in this embodiment), and the resources to be configured. The original dataset is the input dataset for feature engineering (i.e., the multi-dimensional student data set in this embodiment), the training set is the feature set for model training (i.e., the multi-dimensional feature data set in this embodiment), and the test set is the feature set for model training (i.e., the multi-dimensional feature data set in this embodiment). Feature engineering is used for feature selection, and the resources required for algorithm training and testing are configured.

[0090] Step 2: By determining the type, quantity, and distribution of the feature set, determine whether it is necessary to read the entire dataset. If the data volume is too large or the feature distribution is complex, it is not necessary to read the entire dataset, and the sampling process is initiated. If the data volume is small and the feature distribution is balanced, the entire dataset is read directly. Determining whether sample sampling is necessary can be used to set the data structure, storage strategy, and reading process selected in the subsequent algorithm training process, thereby improving the algorithm training speed.

[0091] Step 3: First, determine the sampling rules and extract data. Spark uses RandomForest for feature sampling, involving four samples (1.1, 1.2, 1.3, and 1.4). Then, iterate the sample by setting different parameter combinations (i.e., the optimal algorithm parameter combination for fast selection in this embodiment). Use chain equation interpolation to identify missing data. Simultaneously, for the same data samples (i.e., the samples corresponding to the multi-dimensional feature data set in this embodiment), use a combination of chain equations and Fast Fourier Transform (FFT). Calculate the energy density and set a suitable abnormal energy distribution threshold (i.e., the energy density threshold in this embodiment) to achieve fast convergence of the algorithm. The combination of these two methods completes the identification of missing data and the detection of abnormal data.

[0092] Step 4: During the training process, based on the data processed by the Spark algorithm and the resource load, and in conjunction with the specified Spark Shuffle algorithm rules, select the appropriate Shuffle Manager type to maximize the utilization of machine resources and improve the algorithm training efficiency.

[0093] Step 5: During training, it is necessary to determine whether the model has reached the convergence threshold. If it has not reached the threshold, return to continue iterative training. If the convergence threshold is reached, the algorithm model prediction results can be exported after training is completed. Then, the algorithm model is deployed. At the same time, the PSI index is used to monitor the stability of the prediction. When the PSI index exceeds a specific threshold, the algorithm can be intervened or retrained in a timely manner to ensure the accuracy of detection and prediction of data related to student attrition.

[0094] Step 6: Iterating through all sample sets allows us to derive the optimal parameter configuration combination by adjusting the hyperparameters of the chain equation and the fast Fourier transform. After multiple rounds of adjustments, we can obtain the optimized parameter configuration combination. This combination, along with the exported algorithm model and Spark Shuffle algorithm rules, is then used for subsequent Spark machine resource configuration and deployment. Simultaneously, we continuously monitor the algorithm model's PSI value and make adaptive adjustments. Finally, the process ends.

[0095] Compared with existing technologies, this embodiment improves the accuracy of the target multi-dimensional student dataset and mitigates analytical biases caused by abnormal data fluctuations by determining the initial placeholder values ​​of multi-dimensional feature data samples, constructing a dependency matrix, using a regression model to predict missing values, and merging the imputation results of multi-dimensional feature data subsets. It also enhances the reliability of the target multi-dimensional student dataset by determining the imputation variance data, evaluating the imputation uncertainty data, and performing weighted analysis on the imputation results. Furthermore, it effectively extracts the temporal features of student data by constructing student data sequences based on timestamp information, performing fast Fourier transforms after sliding window segmentation, and determining the frequency domain energy distribution, thus meeting the requirements for data temporal characteristics. Finally, it improves the accuracy of identifying target students in abnormal student states by identifying abnormal frequency domain data sequences with energy density less than a threshold, determining abnormal windows, and performing short-time Fourier transform processing. Finally, it enhances the effectiveness and relevance of the multi-dimensional feature dataset by using the Spark distributed computing framework to extract candidate multi-dimensional feature datasets, merging them, and performing feature selection based on gradient optimization algorithms, adapting to data complexity, and avoiding redundant features interfering with the analysis results.

[0096] Furthermore, as Figure 1 and Figure 2 The specific implementation of the method shown in this embodiment provides a student status analysis device, such as... Figure 5 As shown, the device includes: an acquisition module 31, an identification module 32, and a construction module 33.

[0097] The acquisition module 31 is configured to acquire a multi-dimensional student data set, use the Spark distributed computing framework to extract features from the multi-dimensional student data set, and obtain a multi-dimensional feature data set corresponding to the multi-dimensional student data set. The multi-dimensional student data set includes student enrollment data and student behavior data at school. The identification module 32 is configured to identify missing values ​​in a multi-dimensional feature data set based on a chain equation, obtain missing data in the multi-dimensional feature data set, and perform multiple imputation on the missing data to obtain the target multi-dimensional student data set. Module 33 is configured to construct a student data sequence based on the target multi-dimensional student data set, convert the student data sequence into a frequency domain data sequence, and determine the frequency domain energy distribution of the frequency domain data sequence. The identification module 32 is also configured to identify abnormal data sequences from the student data sequence based on the frequency domain energy distribution, and to perform student status analysis based on the abnormal data sequences to obtain the target students with abnormal student status.

[0098] In some examples of this embodiment, the identification module 32 is specifically configured to: determine the initial placeholder value corresponding to each feature data sample in the multi-dimensional feature data set; construct a dependency matrix corresponding to the multi-dimensional feature data set based on the initial placeholder value; construct a regression model based on the dependency matrix; use the regression model to predict the missing values ​​in the multi-dimensional feature data set to obtain the missing data in the multi-dimensional feature data set; divide the multi-dimensional feature data set into at least one multi-dimensional feature data subset; independently and randomly sample and impute the missing data in at least one multi-dimensional feature data subset to obtain the imputation results corresponding to at least one multi-dimensional feature data subset; and merge the imputation results corresponding to at least one multi-dimensional feature data subset to obtain the target multi-dimensional student data set.

[0099] In some examples of this embodiment, the identification module 32 is further configured to determine the imputation variance data of missing data in at least one multi-dimensional feature data subset during the independent random sampling imputation process, evaluate the imputation uncertainty data corresponding to each of the at least one multi-dimensional feature data subset based on the imputation variance data, and perform weighted analysis on the imputation results corresponding to each of the at least one multi-dimensional feature data subset based on the imputation uncertainty data corresponding to each of the at least one multi-dimensional feature data subset to obtain the target multi-dimensional student data set.

[0100] In some examples of this embodiment, the construction module 33 is specifically configured to: determine the timestamp information of the data in the target multi-dimensional student dataset; convert the target multi-dimensional student dataset into a student data sequence based on the timestamp information; segment the student data sequence based on a sliding window to obtain at least one data sequence segment; perform a fast Fourier transform on the at least one data sequence segment to obtain at least one frequency domain data sequence segment corresponding to the at least one data sequence segment; determine the energy density data corresponding to the at least one frequency domain data sequence segment; and generate the frequency domain energy distribution of the at least one frequency domain data sequence segment based on the energy density data.

[0101] In some examples of this embodiment, the construction module 33 is further configured to select at least one fourth indicator dataset corresponding to the teaching ability dimension from multiple indicator data clusters; perform multi-objective decision analysis on the at least one fourth indicator dataset to determine the fifth parameter data of the target teacher in the teaching effectiveness sub-dimension and the sixth parameter data in the educational outcome sub-dimension; and standardize the fifth parameter data and the sixth parameter data according to a predetermined parameter system to obtain the fourth teaching situation evaluation result.

[0102] In some examples of this embodiment, the construction module 33 is further configured to: identify anomalous frequency domain data sequences with energy densities less than an energy density threshold from the frequency domain energy distribution of at least one frequency domain data sequence segment; determine the anomalous window corresponding to the anomalous frequency domain data sequence; segment the anomalous frequency domain data based on the anomalous window; perform short-time Fourier transform processing on the segmented anomalous data sequence to obtain the time-frequency domain data sequence segment corresponding to the anomalous data sequence segment; determine the anomalous time point and anomalous frequency range corresponding to the anomalous state based on the time-frequency domain data sequence segment; and perform student state analysis based on the anomalous time point and anomalous frequency range to obtain the target student with the anomalous student state.

[0103] In some examples of this embodiment, the acquisition module 31 is specifically configured to acquire a multi-dimensional student data set, use the Spark distributed computing framework to extract features from the multi-dimensional student data set to obtain a candidate multi-dimensional feature data set, and perform feature filtering on the candidate multi-dimensional feature data set according to the gradient optimization algorithm to obtain the multi-dimensional feature data set corresponding to the multi-dimensional student data set.

[0104] It should be noted that other corresponding descriptions of the functional units involved in the student status analysis device provided in this embodiment can be found in [reference needed]. Figure 1 and Figure 2 The corresponding description in [the document] will not be repeated here.

[0105] Based on the above, Figure 1 and Figure 2Accordingly, this embodiment also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method. Figure 1 and Figure 2 The method shown.

[0106] Based on this understanding, the technical solution of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as CD-ROM, USB flash drive, mobile hard drive, etc.) and includes several instructions to cause a computer device (such as personal computer, server, or network device, etc.) to execute the methods of various implementation scenarios of this application.

[0107] like Figure 6 The diagram shown is a hardware structure schematic of an electronic device according to the present invention, comprising: At least one processor 401; and, Memory 402 is communicatively connected to at least one processor 401; wherein, The memory 402 stores instructions that can be executed by at least one processor, such that the at least one processor can perform the student state analysis method as described above.

[0108] Figure 6 Take a processor 401 as an example.

[0109] The electronic device may also include an input device 403 and an output device 404.

[0110] The processor 401, memory 402, input device 403, and output device 404 can be connected via a bus or other means. Figure 6 Taking the example of a connection between China and Israel via a bus.

[0111] Memory 402, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules, such as the program instructions / modules corresponding to the student status analysis method in the embodiments of this application, for example, Figure 1 and Figure 2 The method flow is shown. The processor 401 executes various functional applications and data processing by running non-volatile software programs, instructions, and modules stored in the memory 402, thereby implementing the student status analysis method in the above embodiments.

[0112] Memory 402 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the student status analysis method. Furthermore, memory 402 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, memory 402 may optionally include memory remotely located relative to processor 401, and these remote memories may be connected via a network to the apparatus performing the student status analysis method. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0113] Input device 403 can receive user clicks and generate signal inputs related to user settings and function control for student status analysis methods. Output device 404 may include display devices such as a display screen.

[0114] One or more modules are stored in memory 402, and when run by one or more processors 401, the student status analysis method in any of the above method embodiments is executed.

[0115] Optionally, the aforementioned physical devices may also include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, input units such as a keyboard, etc., and optional user interfaces may also include USB interfaces, card reader interfaces, etc. The network interface may optionally include standard wired interfaces, wireless interfaces (such as Wi-Fi interfaces), etc.

[0116] Those skilled in the art will understand that the physical device structure provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or combine certain components, or have different component arrangements.

[0117] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the aforementioned physical device, supporting the operation of information processing programs and other software and / or programs. The network communication module is used to enable communication between the various components within the storage medium, as well as communication with other hardware and software in the information processing physical device.

[0118] Through the above description of the implementation methods, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platforms, or it can be implemented by hardware. Compared with the existing technology, by applying the solution of this embodiment, this embodiment achieves multi-dimensional feature mining of student enrollment data and on-campus behavior data by acquiring multi-dimensional student datasets and using the Spark distributed computing framework for feature extraction, thus adapting to the needs of data diversity; by identifying missing values ​​and performing multiple imputations on the multi-dimensional feature data set based on chain equations, the integrity of the target multi-dimensional student dataset is improved, solving the problem of missing data from multiple sources; by constructing student data sequences based on the target multi-dimensional student dataset and converting them into frequency domain data sequences, and determining the frequency domain energy distribution, the effective transformation of student data temporal features to frequency domain features is achieved, accurately capturing the correlation information in temporal data; by identifying abnormal data sequences based on the frequency domain energy distribution and performing student status analysis, the accuracy of identifying abnormal student behavior and psychological states in student data with high complexity, diversity, and temporal fusion can be improved, thereby reducing the risk of student attrition; by determining the initial placeholder values ​​of multi-dimensional feature data samples... This study improves the accuracy of the target multi-dimensional student dataset by constructing a dependency matrix, predicting missing values ​​using a regression model, and merging the imputation results of multi-dimensional feature data subsets, thus mitigating analytical biases caused by abnormal data fluctuations. It also enhances the reliability of the target multi-dimensional student dataset by determining the imputation variance data, evaluating imputation uncertainty data, and performing weighted analysis on the imputation results. Furthermore, it effectively extracts temporal features of student data by constructing student data sequences based on timestamp information, performing fast Fourier transforms after sliding window segmentation, and determining the frequency domain energy distribution, thus meeting the requirements for data temporal characteristics. Finally, it improves the accuracy of identifying target students with abnormal student states by identifying abnormal frequency domain data sequences with energy density less than a threshold, determining abnormal windows, and performing short-time Fourier transform processing. Finally, it enhances the effectiveness and relevance of the multi-dimensional feature dataset by using the Spark distributed computing framework to extract candidate multi-dimensional feature datasets, merging them, and performing feature selection based on gradient optimization algorithms, adapting to data complexity, and avoiding redundant features from interfering with the analysis results.

[0119] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0120] The above description is merely a specific embodiment of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments described herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A student status analysis method characterized by, The method comprises the following steps: acquiring a multi-dimensional student data set, performing feature extraction on the multi-dimensional student data set using a Spark distributed computing framework to obtain a multi-dimensional feature data set corresponding to the multi-dimensional student data set, wherein the multi-dimensional student data set comprises student enrollment data and behavior data of students in school; based on a chain equation, identifying missing values in the multi-dimensional feature data set to obtain missing data in the multi-dimensional feature data set, and performing multiple imputations on the missing data to obtain a target multi-dimensional student data set; constructing a student data sequence from the target multi-dimensional student data set, converting the student data sequence into a frequency domain data sequence, and determining the frequency domain energy distribution of the frequency domain data sequence; based on the frequency domain energy distribution, identifying an abnormal data sequence from the student data sequence, and performing student state analysis based on the abnormal data sequence to obtain a target student with an abnormal student state.

2. The method of claim 1, wherein, The method comprises the following steps: determining an initial placeholder value corresponding to each feature data sample in the multi-dimensional feature data set, and constructing a dependency relationship matrix corresponding to the multi-dimensional feature data set based on the initial placeholder value; constructing a regression model based on the dependency relationship matrix, predicting the missing values in the multi-dimensional feature data set using the regression model, and obtaining the missing data in the multi-dimensional feature data set; dividing the multi-dimensional feature data set into at least one multi-dimensional feature data subset, independently and randomly sampling and filling the missing data in the at least one multi-dimensional feature data subset to obtain an imputation result corresponding to each of the at least one multi-dimensional feature data subset; merging the imputation results corresponding to each of the at least one multi-dimensional feature data subset to obtain the target multi-dimensional student data set.

3. The method of claim 2, wherein, The method comprises the following steps: determining the imputation variance data of the missing data in the at least one multi-dimensional feature data subset during the independent random sampling and filling process, and evaluating the imputation uncertainty data corresponding to each of the at least one multi-dimensional feature data subset based on the imputation variance data; based on the imputation uncertainty data corresponding to each of the at least one multi-dimensional feature data subset, performing weighted analysis on the imputation results corresponding to each of the at least one multi-dimensional feature data subset to obtain the target multi-dimensional student data set.

4. The method of claim 1, wherein, The method comprises the following steps: determining the timestamp information of the data in the target multi-dimensional student data set, and converting the target multi-dimensional student data set into the student data sequence based on the timestamp information; segmenting the student data sequence based on a sliding window to obtain at least one data sequence segment, performing fast Fourier transform processing on the at least one data sequence segment to obtain at least one frequency domain data sequence segment corresponding to the at least one data sequence segment; determining energy density data corresponding to the at least one frequency domain data sequence segment, and generating a frequency energy distribution of the at least one frequency domain data sequence segment according to the energy density data.

5. The method of claim 4, wherein, The student state analysis based on the abnormal data sequence comprises: identifying an abnormal frequency domain data sequence with an energy density less than an energy density threshold from the frequency energy distribution of the at least one frequency domain data sequence segment; determining an abnormal window corresponding to the abnormal frequency domain data sequence, segmenting and intercepting the abnormal frequency domain data based on the abnormal window, performing short-time Fourier transform processing on the intercepted abnormal data sequence segment to obtain a time-frequency domain data sequence segment corresponding to the abnormal data sequence segment; determining an abnormal time point and an abnormal frequency range corresponding to an abnormal state according to the time-frequency domain data sequence segment; performing student state analysis according to the abnormal time point and the abnormal frequency range to obtain a target student in an abnormal student state.

6. The method according to any one of claims 1 to 5, characterized in that, The method comprises: obtaining a multi-dimensional student data set, performing feature extraction on the multi-dimensional student data set using a Spark distributed computing framework to obtain a multi-dimensional feature data set corresponding to the multi-dimensional student data set; performing feature screening on the candidate multi-dimensional feature data set according to a gradient optimization algorithm to obtain the multi-dimensional feature data set corresponding to the multi-dimensional student data set.

7. A student state analysis device characterized by comprising: The method comprises: obtaining a multi-dimensional student data set, performing feature extraction on the multi-dimensional student data set using a Spark distributed computing framework to obtain a multi-dimensional feature data set corresponding to the multi-dimensional student data set, wherein the multi-dimensional student data set includes student enrollment data and student behavior data in school; performing missing value identification on the multi-dimensional feature data set based on a chain equation to obtain missing data in the multi-dimensional feature data set, and performing multiple imputation on the missing data to obtain a target multi-dimensional student data set; constructing a student data sequence according to the target multi-dimensional student data set, converting the student data sequence into a frequency domain data sequence, and determining a frequency energy distribution of the frequency domain data sequence; identifying an abnormal data sequence from the student data sequence according to the frequency energy distribution, and performing student state analysis based on the abnormal data sequence to obtain a target student in an abnormal student state.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by a processor to implement the method of any one of claims 1 to 6. The computer program is executed by a processor to implement the method of any one of claims 1 to 6.

9. An electronic device comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, characterized in that, The computer program is executed by the processor to implement the method of any one of claims 1 to 6.

10. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the method of any one of claims 1 to 6.

Images