Student performance prediction method based on hybrid deep learning and attention mechanism

Abstract

The invention relates to the technical field of machine learning, deep learning, data mining and the like, in particular to a student performance prediction method based on hybrid deep learning and an attention mechanism. On-line learning log information of students is utilized, preprocessing and feature extraction are firstly carried out on the log information, and a correlation feature selection method is used to obtain features having important influence on a prediction result. And then weekly accumulative statistics and all accumulative statistics are carried out on the extracted features according to feature categories. Then, the extracted time sequence features and potential features are spliced, and due to the fact that the difference between the two types of learned features is large in magnitude and category, a deep neural network is used for high-order feature interaction, and deeper features are learned; at the same time, different influence degrees of different features on student performance are considered, and an attention mechanism is used to distribute different weights for different deep-level features. And finally, a sigmoid classifier is used to predict whether the student can pass a certain course, thereby improving the prediction accuracy.

Description

(1) Technical field [0001] The invention relates to technical fields such as machine learning, deep learning and data mining, and in particular to a method for predicting student performance based on hybrid deep learning and attention mechanism. (2) Background technology [0002] In recent years, education informatization has developed rapidly, and the wide application of distance education, online learning and other systems has provided a large amount of rich data for educational data mining, making educational data mining usher in a new turning point. Using these data, researchers can identify students' learning environment, learning conditions, and learning status, realize the explanation of some educational phenomena, and improve the effectiveness of education. Among them, student performance prediction, as one of the important fields of educational data mining, has received more and more attention. By predicting student performance, students with learning risks can be ...

AI Technical Summary

Problems solved by technology

However, in the field of student performance prediction, there are three obvious shortcomings in the existing research based on recurrent neural network and its variant methods: (1) In the data processing stage, only relying on the ability of the model itself to process features processing, lack of effective feature extraction process

(2) Long-short-term memory neural network (LSTM) is mostly used. Although this method can learn longer time series data compared with the cyclic neural network, its structure is complex and the number of parameters is large, especially when the LSTM is extended to a deep layer. When , the number of parameters increases sharply, which makes the model prone to overfitting and slows down the training speed.

Images