Predicting personalized academic performance by collaborative filtering algorithm

In the era of digital intelligence, large-scale educational data has been retained． These digital resources make personalized education feasible． It is challenging to evaluate individual student ability precisely from the vast data by program, and to provide proper exercises to make the learning process effective and sustainable. In this paper, the score data of each student on each question were subject to analysis by a collaborative filtering recommendation algorithm to predict student academic scores. Accuracy of prediction was evaluated after comparison of predicted scores and actual scores． Robustness of prediction was assessed after examination of crucial parameters． Threshold of similarity was found a key factor for prediction accuracy． But high threshold weakened prediction performance for abnormal cases． Sample size exhibited strong robustness for multi-subject academic performance prediction． Keeping lifelong learning and international perspective in mind, our findings provide an applicable method for prediction, which is crucial for personalized learning and development of intelligent educational system．