A Study of Academic Achievement Attribution Analysis Based on Explainable Machine Learning Techniques

The research on academic achievement prediction and attribution is a frontier topic in the field of artificial intelligence education. However, because most artificial intelligence algorithms are deep and non-parametric, it is difficult for the algorithm itself to give a credible explanation of the prediction and attribution results, which is very dangerous for the education of cultivating students with rich emotions and values. Explainable AI technology can assist educational researchers to clarify the decision-making process of the algorithm, help researchers accurately locate the key factors affecting academic achievement, and formulate targeted learning intervention strategies. Based on the open source data set of the Advanced Innovation Center for Future Education (AICFE) of Beijing Normal University, this research constructs a feature set based on Weiner's attribution theory, and finally determines the naive Bayes algorithm as the best prediction algorithm through feature selection, model training, and model selection; Then, using machine learning interpretability framework SHAP, we explored the factors that affect students' academic achievements from the group and individual levels respectively. The group-level exploration is helpful to reveal the overall impact trend of various influencing factors on academic achievements, and help to formulate group-oriented learning intervention strategies; The exploration at the individual level is conducive to finding the personalized factors that affect academic achievement and achieving targeted intervention.