Predicting School Children Academic Performance Using Machine Learning Techniques

Abstract

The study aims to assess the machine learning techniques in predicting students' associated factors that affect their academic performance. The study sample consisted of 5084 middle and high school students between the ages of 10 and 17, attending public and UNRWA schools in the West Bank. The 'Health Behaviors School Children' questionnaire for the 2013-2014 academic year was used for data collection, and was then analyzed through machine learning techniques in order to evaluate their relationship with student academic outcomes. Six machine learning techniques (Random Forest, Neural Network, Support Vector Machine, Decision Tree, Nave Bayes, and Logistic Regression) were used for prediction. The results indicated that the logistic regression and Nave Bayes models had the highest accuracy levels (94.3%, 94%) respectively, followed by a decision tree, Neural Network, Random Forest, and Support Vector Machine (93.3%,91.9%,91.7%, and 80.2%) respectively. Thus, the Logistic Regression and Nave Bayes had the best performance in classifying and predicting student academic performance with the associated factors. Furthermore, Decision Tree, Random Forest, and Neural Network had better predictive performance than Support Vector Machine. The results indicated that perception, Smoking, Depression, PTSD, Healthy Food Consumption, Age, gender, Grade Level, and Family income are the most important and significant factors that influence student academic performance. Overall, machine learning techniques prove efficient tools for identifying and predicting the features that influence student academic performance. The deployment of machine learning techniques within schools' information systems will facilitate the development of health prevention and intervention programs that will enhance students' academic performance.