Predictive Modeling of Student Academic Outcomes Through Feature-Engineered Supervised Learning

Abstract

To receive proper help and effective educational planning, one must predict the academic results of the pupils. In this work, the machine learning approach is applied to research the key factors that influence academic performance of students, 17 features that include demographic data, behavioral (raising hands, visiting resources, watching announcements, and participating in discussions) and parental involvement (survey participation and school satisfaction) data, and attendance records of 480 students were analyzed. The students were categorized as three groups namely: High (H), Medium (M), and Low (L), according to their performance. Random Forest was selected as the best classification model after testing various other classifier models and the optimized model gave the best classification accuracy of 79.17% In order to resolve the uneven performance distribution, this model was set with the estimators numbered 600, depth to its maximum of 20 and the weights of the classes were equal. The following behaviors were identified to be significant contributors, student engagement behavior, parental satisfaction, educational stage, and absence patterns. The research proves that machine learning can be successfully used to predict academic achievement and help teachers to recognize at-risk students and intervene in their areas of need. The presented piece of work provides a handy reference to developing the performance prediction systems of students and fits in the growing body of research in the area of the educational data mining.