Role of Hybrid Feature Selection Algorithms in Foreseeing Student Performance

The interest of analysts in applying analytical algorithms to investigate students’ performance data with a view to enhancing their knowledge is growing among data miners. The key factor of this trend is ample multimedia data generated by educational institutions with the usage of technologies, tools of e-learning, and other digital platforms for conducting online courses. Educators could utilize these data to examine and understand students’ learning behaviors by using data mining techniques to forecast student achievement, among other things. In data mining models, the difficult task is to select effective strategies that will generate satisfactory forecast accuracy. To achieve such goals, this article uses six hybrid feature selection (FS) algorithms such as hybrid PSO, hybrid GA, filter wrap, wrapper embedded, ensemble based, and filter embedded and five algorithms such as “Random Forest (RF)”, “Decision Tree (DT)”, “Logistic Regression (LR)”, “K-Nearest Neighbor (KNN)”, and “Support Vector Machine” (SVM) to forecast the performance of students and to make comparison of performance of five classification processes in the perspective of six FS models. Based on the analysis of the given data set, it is concluded that the performance of classification models improved when 20 important features were part of the data in place of all 30 attributes. Further reduction in the number of attributes did not result in further improvement in the performance of classification models in predicting students’ success or failure.