Score prediction method and system based on improved attention mechanism
By improving the attention mechanism and residual enhancement connection neural network model, and combining it with an adaptive momentum decay optimizer, the problems of insufficient data mining depth and detection accuracy in student performance prediction models are solved, and accurate classification and efficient prediction of students' learning status are achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-18
- Publication Date
- 2026-03-10
AI Technical Summary
Existing student performance prediction models have shortcomings in terms of data mining depth and detection accuracy, especially the slow convergence speed and low detection accuracy of neural network-based algorithms.
We employ an improved attention mechanism and a neural network model with residual reinforcement connections, combined with an adaptive momentum decay AdamW optimizer. Through principal component analysis dimensionality reduction and class weight embedding encoding, we construct an improved class weight encoding neural network model. We also add attention mechanism layers between hidden layers and use a temperature-regulated softmax activation function to optimize feature representation and prevent overfitting.
It improved the precision, recall, and F1 score of student performance prediction, achieved accurate classification of student learning status, and enhanced the model's learning performance and prediction accuracy.
Smart Images
Figure CN121638535A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of performance data analysis and processing technology, and in particular relates to a performance prediction method and system based on an improved attention mechanism. Background Technology
[0002] With the continuous development of artificial intelligence and big data technologies, student performance prediction is receiving increasing attention in the fields of artificial intelligence and educational data mining (EDM). Although many organizations in academia recognize the importance of leveraging the potential of digital transformation to enhance the educational process, and many universities are already actively exploring data-driven smart teaching, overall, the value of data mining is not yet high enough, resulting in a large amount of effective data being buried and making it difficult to support the in-depth development of comprehensive digital transformation in education.
[0003] Currently, numerous machine learning algorithms exist for predicting student grades, such as traditional methods like linear regression, decision trees, and support vector machines. However, these methods often require manual feature selection and struggle to uncover deeper information within the data. Dorrer et al. proposed building models for predicting future grades using neural network regression tools. They cross-validated the datasets on both the training and test sets, using the coefficient of determination and root mean square error (RMSE) to evaluate the model's predictive ability. Deep learning-based grade prediction models, on the other hand, can more fully mine the implicit information within deeper structures. Liu Tong et al. proposed a flexible combination of LSTM and attention mechanisms to simultaneously extract features from grade sequence information across both course and time dimensions. Addressing the issue of insufficient training data in early academic years, they constructed a method for calculating similar students based on temporal co-occurrence frequency, integrating information from similar students to achieve information complementarity.
[0004] It is evident that different network models have been widely applied in student performance prediction tasks. Although neural network algorithms have good learning performance, some neural network-based algorithms still suffer from problems such as slow convergence speed and low detection accuracy. Summary of the Invention
[0005] The present invention aims to solve the above problems and provide a performance prediction method and system based on an improved attention mechanism.
[0006] In a first aspect, the present invention provides a performance prediction method based on an improved attention mechanism, comprising the following steps: Step 1: Obtain the raw score data and preprocess it; divide the preprocessed data into a test set and a training set; Step 2: Calculate the correlation coefficient matrix between each column in the dataset and perform a correlation analysis on all formative achievements; Step 3: Extract data features using a support vector classifier; perform principal component analysis for dimensionality reduction to obtain the dimensionality-reduced features of the training and test sets. Step 4: Construct a neural network model with improved category weight encoding, including an input layer, two hidden layers, an attention mechanism layer, and an output layer; The input layer takes in the dimensionality-reduced data features; each hidden layer contains a convolutional layer and a fully connected layer; the fully connected layer uses the ReLU activation function and is followed by a Dropout layer. An improved attention mechanism layer is added between the aforementioned hidden layers to weight the features; Calculate attention score: e i = tanh ( W a h i + b a );in, h i These are input features. Wa It is a weight matrix. b a It is a bias term; Calculate attention weights: in, N It is the number of input features; Multiply the input features by the attention weights: ; Introducing residual-enhanced connectivity: ;in, FC Indicates a fully connected layer; The output layer uses a temperature-regulated type. soft Max activation function: ; in, y =( y 1 , y 2 ,… y n ) represents the input vector. ; T This represents a temperature parameter, which can take any value. T When =1, the output distribution is as follows: soft max, T The larger the value, the smoother the output distribution. If you want a greater degree of randomness in the output distribution, you need to increase the temperature parameter. T The value of ; Step 5: Train the model; introduce the adaptive momentum decay AdamW optimizer: ; in w t This is the current parameter. m t and v t These are the first and second moments of the gradient, respectively. ò It is a small constant. λ It is the momentum decay factor. It is the learning rate; this method can decouple weight decay from gradient update, and can more stably "suppress large parameters" during training, making parameter updates smoother and effectively reducing the risk of overfitting; The model is trained based on the dimensionality-reduced features; for each training round, the log loss is calculated using the class weights of the training set, the embedded encoded labels, and the output, and backpropagation is performed to calculate the gradient to update the weights and biases of the neural network. ; in, y It is the category weight embedding encoding of the real label. It is the probability distribution predicted by the model. N It is the sample size. K It is the number of categories; Step 6: Model evaluation; Based on the model trained in step 5, make predictions on the dimensionality-reduced test set data and obtain the predicted label for each test sample.
[0007] Furthermore, in the performance prediction method based on the improved attention mechanism described in this invention, the preprocessing in step 1 includes the following steps: Step 1.1: Remove blank data; Step 1.2: Calculate the residual sequence of all data, and use the root mean square of the sum of squared residuals as the standard deviation. If the absolute value of the residual is greater than twice the standard deviation, then correct the outlier data. Step 1.3: Divide the final exam scores in the training data into score levels and label them as 5 label arrays; convert the label arrays into the encoding form of class weight embedding: first calculate the proportion of each label sample and generate a class weight vector, where the weight vector is normalized by taking the inverse of the sample proportion; then map the labels into low-dimensional dense vectors and multiply the vectors element by element with the class weights.
[0008] Furthermore, in the performance prediction method based on the improved attention mechanism described in this invention, the test set and training set data are divided in a ratio of 0.2:0.8.
[0009] Furthermore, in the performance prediction method based on the improved attention mechanism described in this invention, when dividing the final exam scores in the training data according to score levels, the bottom 5% are marked as label 0; the bottom 25% are marked as label 1; the bottom 50% are marked as label 2; the top 5% are marked as label 3; and scores outside the above ranges are marked as label 4.
[0010] Furthermore, in the performance prediction method based on the improved attention mechanism described in this invention, step 3, principal component analysis, includes: principal components are obtained by eigenvalue decomposition of the covariance matrix of the data; let the data matrix be... X The sample size is n Its covariance matrix is C , T To represent the transpose operation, then: ; right C Eigenvalue decomposition: C = V Λ V T in, V It is an eigenvector matrix, Λ is a diagonal matrix, and its diagonal elements are eigenvalues; select the first... k The eigenvectors corresponding to the largest eigenvalues form the projection matrix. W Projecting data into a lower-dimensional space: X pca = XW .
[0011] Furthermore, in the performance prediction method based on the improved attention mechanism described in this invention, step 6, predicting the dimensionality-reduced test set data, specifically includes: The accuracy, precision, recall, and F1 score of the model are calculated using the accuracy_score, precision_score, recall, and f1_score functions, and the results are printed out. ; in, TP For a true example, the actual result is positive and the prediction is positive; FP This is a false positive, meaning the actual value is negative but the prediction is positive. FN This is a false counterexample, meaning the actual value is positive but the prediction is negative; TN A true counterexample is one where the actual value is negative and the prediction is negative. ; in, Precision Precision rate refers to the proportion of samples predicted as positive that actually belong to the positive class.Recall Recall rate refers to the proportion of samples that are actually positive that are correctly predicted. F 1 represents F The 1-score, which is the harmonic mean of precision and recall, is used to comprehensively evaluate the performance of both, and they are both commonly used evaluation metrics in classification tasks.
[0012] In a second aspect, the present invention provides a performance prediction system based on an improved attention mechanism, including a data preprocessing module, a dimensionality reduction module, a model building module, and a model training and output module; The data preprocessing module is used to acquire the raw performance data and preprocess it; divide the preprocessed data into a test set and a training set; calculate the correlation coefficient matrix between each column in the dataset and perform correlation analysis on all formative scores; The dimensionality reduction module is used to extract data features using a support vector classifier; perform principal component analysis dimensionality reduction to obtain the dimensionality reduction features of the training set and the test set; The model building module is used to build a neural network model with improved category weight encoding; The model training and output module is used to train the model based on the dimensionality-reduced features. For each round of training, the class weights of the training set are used to embed the encoded labels and output to calculate the log loss, and backpropagation is performed to calculate the gradient to update the weights and biases of the neural network. At the same time, the trained model is used to predict the dimensionality-reduced test set data and output the predicted label for each test sample.
[0013] Furthermore, the performance prediction system based on the improved attention mechanism described in this invention includes an input layer, two hidden layers, an attention mechanism layer, and an output layer in the improved category weight encoding neural network model. The input layer takes in the dimensionality-reduced data features; each hidden layer contains a convolutional layer and a fully connected layer; the fully connected layer uses the ReLU activation function and is followed by a Dropout layer. An improved attention mechanism layer is added between the aforementioned hidden layers to weight the features; Calculate attention score: e i = tanh ( W a h i + b a );in, h i These are input features. Wa It is a weight matrix. b a It is a bias term; Calculate attention weights: ;in, N It is the number of input features; Multiply the input features by the attention weights: ; Introducing residual-enhanced connectivity: ;in, FC Indicates a fully connected layer; The output layer uses a temperature-regulated type. soft Max activation function: ; in, y =( y 1 , y 2 ,… y n ) represents the input vector. . T This represents a temperature parameter, which can take any value. T The larger the value, the smoother the output distribution.
[0014] Thirdly, the present invention provides an electronic device for performance prediction based on an improved attention mechanism, comprising a processor and a memory electrically connected to each other; the memory is used to store a computer program; when the processor executes the aforementioned computer program, it can implement the performance prediction method based on the improved attention mechanism described in the first aspect.
[0015] Fourthly, the present invention provides a computer-readable storage medium storing a computer program thereon; when the computer program is executed, it can implement the performance prediction method based on the improved attention mechanism described in the first aspect.
[0016] The performance prediction method and system based on the improved attention mechanism described in this invention effectively preserves key information about performance data classification thinking by adding an attention mechanism layer between the hidden layers of the network and introducing residual enhancement connections when constructing the network model. At the same time, a loss function module based on the embedding encoding of real label category weights is designed, and a Dropout layer is added to prevent overfitting and optimize feature representation. Attached Figure Description
[0017] Figure 1 This is a schematic diagram of the performance prediction method based on the improved attention mechanism described in an embodiment of the present invention. Detailed Implementation
[0018] The following detailed description of the performance prediction method and system based on the improved attention mechanism described in this invention is provided with reference to the accompanying drawings and embodiments.
[0019] Example 1 This embodiment discloses a performance prediction method based on an improved attention mechanism, such as... Figure 1 As shown, it includes the following steps: Step 1: Obtain the raw score data and preprocess it; divide the preprocessed data into a test set and a training set; In this embodiment of the disclosure, the test set and training set data are divided in a ratio of 0.2:0.8; the preprocessing includes the following steps: Step 1.1: Remove blank data; Step 1.2: Calculate the residual sequence of all data, and use the root mean square of the sum of squared residuals as the standard deviation. If the absolute value of the residual is greater than twice the standard deviation, then correct the outlier data. Step 1.3: Divide the final exam scores in the training data into score levels and label them as 5 label arrays; convert the label arrays into the encoding form of class weight embedding: first calculate the proportion of each label sample, and generate the class weight vector, where the weight vector is normalized by taking the inverse of the sample proportion; In this embodiment of the disclosure, when dividing the final exam scores in the training data according to score levels, the bottom 5% are labeled as label 0; the bottom 25% are labeled as label 1; the bottom 50% are labeled as label 2; the top 5% are labeled as label 3; and scores outside the above ranges are labeled as label 4.
[0020] The labels are then mapped to low-dimensional dense vectors, and the vectors are multiplied element-wise with the class weights. In this embodiment, the vector dimension is set to 8, and the vectors are multiplied element-wise with the class weights. This encoding preserves class discrimination and guides the model to focus on key levels with a small sample size, such as the top 5% and bottom 5%, which are precisely the student performance levels that should be focused on. Furthermore, it can directly update the embedding vector through gradients during backpropagation, enhancing the model's understanding of class semantics.
[0021] Step 2: Calculate the correlation coefficient matrix between each column in the dataset and perform a correlation analysis on all formative achievements; Step 3: Extract data features using a support vector classifier (SVC), with parameter C set to 9.5 and gamma parameter set to scale; perform principal component analysis (PCA) dimensionality reduction, retaining 90% of the variance, to obtain the dimensionality-reduced features of the training and test sets; In this embodiment of the disclosure, the principal components are obtained by performing eigenvalue decomposition on the covariance matrix of the data; let the data matrix be... X The sample size is n Its covariance matrix is C , T To represent the transpose operation, then: ; right C Eigenvalue decomposition: C = V Λ V T in, V It is an eigenvector matrix, Λ is a diagonal matrix whose diagonal elements are eigenvalues; select the first... k The eigenvectors corresponding to the largest eigenvalues constitute the projection matrix. W Projecting data into a lower-dimensional space: X pca = XW .
[0022] Step 4: Construct a neural network model with improved category weight encoding, including an input layer, two hidden layers, an attention mechanism layer, and an output layer; The input layer takes in the dimensionality-reduced data features; each hidden layer contains a convolutional layer and a fully connected layer; the fully connected layer uses the ReLU activation function and is followed by a Dropout layer. An improved attention mechanism layer is added between the aforementioned hidden layers to weight the features; Calculate attention score: e i = tanh ( W a h i + b a );in, h i These are input features. Wa It is a weight matrix. b a It is a bias term; Calculate attention weights: ;in, N It is the number of input features; Multiply the input features by the attention weights: ; Introducing residual-enhanced connectivity: ;in, FC Indicates a fully connected layer; The output layer uses a temperature-regulated type. soft The max activation function is used for multi-class classification tasks. ; in, y =( y 1 ,y 2 ,… y n ) represents the input vector. . T This represents a temperature parameter, which can take any value. T The larger the value, the smoother the output distribution.
[0023] In this embodiment of the disclosure, a temperature coefficient is set. T The initial value is 1. T Updates are performed via backpropagation. This is especially important when the model has high uncertainty regarding its predictions. T The value will adaptively increase, which reduces the sharpness of the probability distribution and avoids overconfident predictions. And when the model clearly distinguishes between categories, T It will adaptively decrease in size to enhance the probability proportion of different categories and improve prediction accuracy.
[0024] Step 5: Train the model; introduce the adaptive momentum decay AdamW optimizer: ; in w t This is the current parameter. m t and v t These are the first and second moments of the gradient, respectively. ò It is a small constant. λ It is the momentum decay factor. It is the learning rate; in this embodiment of the disclosure, a momentum decay factor is set. λ The initial value is 0.95, and the learning rate is set. The value is 0.01, and the momentum parameter is adjusted every 5 epochs. w t Perform a decay update; The model is trained based on the dimensionality-reduced features, with 50 epochs and each batch containing 64 samples. For each training round, the log loss is calculated using the class weights of the training set, embedded encoding labels, and outputs, and backpropagation is performed to calculate gradients to update the weights and biases of the neural network. ; in, y It is the category weight embedding encoding of the real label. It is the probability distribution predicted by the model. N It is the sample size. K It represents the number of categories.
[0025] Step 6: Model evaluation; Based on the model trained in step 5, make predictions on the dimensionality-reduced test set data and obtain the predicted label for each test sample.
[0026] In this embodiment of the disclosure, the prediction of the dimensionality-reduced test set data specifically includes: The accuracy, precision, recall, and F1 score of the model are calculated using the accuracy_score, precision_score, recall, and f1_score functions, and the results are printed out. ; in, TP For a true example, the actual result is positive and the prediction is positive; FP This is a false positive, meaning the actual value is negative but the prediction is positive. FN This is a false counterexample, meaning the actual value is positive but the prediction is negative; TN A true counterexample is one where the actual value is negative and the prediction is negative. ; in, Precision Precision rate refers to the proportion of samples predicted as positive that actually belong to the positive class. Recall Recall rate refers to the proportion of samples that are actually positive that are correctly predicted. F 1 represents F The 1-score, which is the harmonic mean of precision and recall, is used to comprehensively evaluate the performance of both, and they are both commonly used evaluation metrics in classification tasks.
[0027] This embodiment can achieve accurate classification and prediction of students' learning status. It uses artificial intelligence algorithms such as BP neural network, K-nearest neighbor algorithm, random forest and support vector machine to perform detailed classification and prediction for students' grades in different intervals such as the top 5%, bottom 5%, bottom 25% and bottom 50%. Among them, label 0 represents the bottom 5%, label 1 represents the bottom 25%, label 2 represents the bottom 50%, label 3 represents the top 5%, and label 4 represents those not in the above range.
[0028] Example 2 Based on the above embodiment 1, this embodiment discloses a performance prediction system based on an improved attention mechanism, including a data preprocessing module, a dimensionality reduction module, a model building module, and a model training and output module; The data preprocessing module is used to acquire the raw performance data and preprocess it; divide the preprocessed data into a test set and a training set; calculate the correlation coefficient matrix between each column in the dataset and perform correlation analysis on all formative scores; The dimensionality reduction module is used to extract data features using a support vector classifier; perform principal component analysis dimensionality reduction to obtain the dimensionality reduction features of the training set and the test set; The model building module is used to build a neural network model with improved category weight encoding; The model training and output module is used to train the model based on the dimensionality-reduced features. For each round of training, the class weights of the training set are used to embed the encoded labels and output to calculate the log loss, and backpropagation is performed to calculate the gradient to update the weights and biases of the neural network. At the same time, the trained model is used to predict the dimensionality-reduced test set data and output the predicted label for each test sample.
[0029] In this embodiment of the disclosure, the improved category weight encoding neural network model includes an input layer, two hidden layers, an attention mechanism layer, and an output layer; The input layer takes in the dimensionality-reduced data features; each hidden layer contains a convolutional layer and a fully connected layer; the fully connected layer uses the ReLU activation function and is followed by a Dropout layer. An improved attention mechanism layer is added between the aforementioned hidden layers to weight the features; Calculate attention score: e i = tanh ( W a h i + b a );in, h i These are input features. Wa It is a weight matrix. b a It is a bias term; Calculate attention weights: ;in, N It is the number of input features; Multiply the input features by the attention weights: ; Introducing residual-enhanced connectivity: ;in, FC Indicates a fully connected layer; The output layer uses a temperature-regulated type. soft Max activation function: ; in, y =( y 1 , y 2 ,… yn ) represents the input vector. . T This represents a temperature parameter, which can take any value. T The larger the value, the smoother the output distribution.
[0030] The specific operation steps of the performance prediction system based on the improved attention mechanism described in this embodiment are the same as those of the performance prediction method based on the improved attention mechanism described in Embodiment 1 above, and will not be repeated here.
[0031] Example 3 This embodiment discloses an electronic device for performance prediction based on an improved attention mechanism, including a processor and a memory electrically connected to each other; the memory is used to store a computer program; when the processor executes the aforementioned computer program, it can implement the performance prediction method based on the improved attention mechanism as described in Embodiment 1. The specific prediction method steps are the same as those in Embodiment 1, and will not be repeated here.
[0032] Example 4 This embodiment discloses a computer-readable storage medium storing a computer program. When the computer program is executed, it can implement the performance prediction method based on the improved attention mechanism as described in Embodiment 1. The specific prediction method steps are the same as those in Embodiment 1, and will not be repeated here.
[0033] The computer described in this application embodiment can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. The computer-readable storage medium can be any usable medium that a computer can read, or a data storage device such as a server or data center that integrates one or more usable media. The usable medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., digital versatile optical disc (DVD)), or a semiconductor medium (e.g., solid-state drive (SSD)). The software formed by the computer's stored code can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other storage media that are mature in the art.
[0034] In the various embodiments of this application, the functional modules can be integrated into one processing unit or module, or each module can exist physically separately, or two or more modules can be integrated into one unit or module. In the above embodiments, they can be implemented entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, they can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated.
[0035] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A performance prediction method based on an improved attention mechanism, characterized in that, It comprises the following steps: Step 1, obtaining original score data and preprocessing; Divide the preprocessed data into a test set and a training set; Step 2, calculate the correlation coefficient matrix between each column in the data set; Step 3, extract data features using a support vector classifier; Perform principal component analysis dimensionality reduction operation to obtain the dimensionality reduction features of the training set and the test set; Step 4, build a neural network model with class weight encoding, including an input layer, two hidden layers, an attention mechanism layer, and an output layer; The input layer inputs the dimensionality reduction processed data features; each hidden layer contains a convolution layer and a fully connected layer; the fully connected layer uses a Relu activation function, and a Dropout layer is added after it; Add an improved attention mechanism layer between the aforementioned hidden layers to weight the features; Computing attention scores: e i = tanh ( W a h i + b a ); wherein, h i is an input feature, Wa is a weight matrix, b a is a bias term; Computing attention weights: wherein, N is the number of input features; Multiply the input features and the attention weights: ; Introducing a residual enhancement connection: ; wherein, FC denotes a fully connected layer; The output layer uses temperature regulation type soft max activation function: ； wherein y ( y 1 , y 2 ,… y n ) denotes an input vector, ; T denotes a temperature parameter; Step 5, train the model; introduce the AdamW optimizer with adaptive momentum decay: ； wherein w t is the current parameter, m t and v t are the first and second moment estimates of the gradient, respectively, ò is a small constant, λ is a momentum decay factor, is the learning rate; Train the model based on the dimensionality reduced features; for each training round, use the class weight embedding encoding label and output to calculate the log loss, and perform back propagation to calculate the gradient to update the weights and biases of the neural network; ； wherein, y is a class weight embedding encoding of the true label, is a probability distribution predicted by the model, N is the number of samples, K is the number of classes; Step 6, model evaluation; based on the model trained in step 5, predict the dimensionality reduced test set data and obtain the predicted label of each test sample. 2.The achievement prediction method based on the improved attention mechanism according to claim 1, characterized in that, The preprocessing of step 1 comprises the following steps: Step 1.1, remove blank data; Step 1.2, calculate the residual sequence of all data, use the root mean square of residual sum of squares as the standard deviation, if the absolute value of residual is greater than twice the standard deviation, modify the abnormal data; Step 1.3, divide the final score in the training data according to the score level and mark it as 5 label arrays; convert the label array to class weight embedding encoding form: first calculate the proportion of each label sample to generate a class weight vector, where the weight vector is the inverse of the sample proportion after normalization; then map the label to a low-dimensional dense vector and multiply the vector and the class weight element by element. 3.The achievement prediction method based on the improved attention mechanism according to claim 2, characterized in that, The test set and the training set are divided according to the ratio of 0.2:0.
8. 4.The method of claim 2, wherein, When dividing the final score in the training data according to the score level, the last 5% is marked as label 0; the last 25% is marked as label 1; the last 50% is marked as label 2; the first 5% is marked as label 3; the scores not in the above interval are marked as label 4. 5.The method of claim 1, wherein, The principal component analysis described in step 3 includes: the principal components are obtained by eigen decomposition of the covariance matrix of the data; let the data matrix be X , the number of samples be n , the covariance matrix of the data be C , T denote the transposition operation, then: ； To C Eigen decomposition: C = V Λ V T wherein, V is the eigenvector matrix, Λ is a diagonal matrix whose diagonal elements are the eigenvalues; select the eigenvectors corresponding to the first k largest eigenvalues to form the projection matrix W , project the data to a low-dimensional space: X pca = XW 。 6.The method of claim 1, wherein, Step 6, the prediction of the dimensionality reduced test set data specifically comprises: Use the accuracy_score, precision_score, recall_score and f1_score functions to calculate the accuracy, precision, recall and F1 score of the model to evaluate the model, and print the results; ； wherein, TP is a true positive, i.e. actually positive predicted positive; FP is a false positive, i.e. actually negative but predicted positive; FN is a false negative, i.e. actually positive but predicted negative; TN is a true negative, i.e. actually negative predicted negative; ； wherein, Precision denotes precision; Recall denotes recall; F 1 denotes F 1 score refers to the harmonic mean of precision and recall.
7. A performance prediction system based on improved attention mechanism, characterized in that: It comprises a data preprocessing module, a dimensionality reduction operation module, a model building module, a model training and output module; The data preprocessing module is used to obtain original score data and preprocess it; divide the preprocessed data into a test set and a training set; Correlation analysis is performed on all formative achievements by calculating the correlation coefficient matrix between columns in the data set; The dimension reduction operation module is configured to extract data features by using a support vector classifier; The principal component analysis dimension reduction operation is performed to obtain dimension reduced features of the training set and the test set; The model construction module is configured to construct a neural network model with improved class weight encoding; The model training and output module is configured to train the model based on the dimension reduced features; For each round of training, the class weight embedding is used to encode the label and output to calculate the log loss, and back propagation is performed to calculate the gradient to update the weights and biases of the neural network; at the same time, the trained model is used to predict the dimension reduced test set data, and the prediction label of each test sample is output. 8.The achievement prediction system based on the improved attention mechanism of claim 7, wherein: The neural network model with improved class weight encoding includes an input layer, two hidden layers, an attention mechanism layer and an output layer; The input layer inputs the dimension reduced data features; each hidden layer includes a convolution layer and a fully connected layer; the fully connected layer uses a Relu activation function, and a Dropout layer is added after it; An improved attention mechanism layer is added between the aforementioned hidden layers to weight the features; Computing attention scores: e i = tanh ( W a h i + b a ); wherein, h i is an input feature, Wa is a weight matrix, b a is a bias term; Computing attention weights: ; wherein N is the number of input features; Multiply the input features and the attention weights: ; Introducing a residual enhancement connection: ; wherein, FC denotes a fully connected layer; The output layer uses temperature regulation type soft max activation function: ； wherein y ( y 1 , y 2 ,… y n ) denotes an input vector, ; T denotes a temperature parameter.
9. An achievement prediction electronic device based on an improved attention mechanism, comprising a processor and a memory electrically connected; the memory is used to store a computer program; characterized in that: When the processor executes the aforementioned computer program, the achievement prediction method based on the improved attention mechanism according to any one of claims 1-6 can be implemented.
10. A computer readable storage medium having stored thereon a computer program; characterized in that: When the computer program is executed, the achievement prediction method based on the improved attention mechanism according to any one of claims 1-6 can be implemented.