Predicting and Analyzing College Students’ Performance Based on Multifaceted Data Using Machine Learning

During the teaching process of college courses, prediction of students' final performance at early stages can help teachers intervene students and improve the teaching effects. In recent years, there have been some researches on predicting students' performance based on machine learning techniques. However, many existing works lack comprehensive and sufficient student data, and there is still room for improvement in the effectiveness of predictions. Moreover, many existing methods require obtaining student data of the whole semester, which are unavailable until the end of the semester. In order to alleviate the above problems, 576 students' data are collected during five years teaching of our Software Testing course. The multifaceted data consists of 39 attributes covering students' demographic information, theoretical learning activities, practical training activities and contest learning activities. Several prediction models are created based on logical regression, random forest, and convolutional neural network. Experimental studies show that with the student data available in the middle or second half of the semester, our models can effectively predict students' final performance. Furthermore, the student features that most affect the prediction results are analyzed in the experiments.