Prediction of Students Academic Performance Utilizing Hybrid Teaching-Learning based Feature Selection and Machine Learning Models

Students' performance prediction problem is a challenging task that faces educational systems every semester. The educational systems have a large amount of data that can help tutors build robust strategies to enhance the educational systems. This paper proposed a robust hybrid method between a set of machine learning (ML) methods and a wrapper feature selection method. ML methods are used as classification methods, while Binary Teaching-Learning Based Optimization (TLBO) is used as a feature selection algorithm. Two real datasets are used in this paper. The obtained results reveal that Logistic Regressing (LR) and Linear Discriminant Analysis (LDA) achieve the best classification performance. Moreover, the TLBO proves its ability to enhance the overall performance of ML algorithms. By testing TLBO with LDA, the AUC result is improved up to 3% and 8% for both tested datasets compared to LDA result without TLBO.