Feature Engineering for Students Academic Performance and Reignition using Machine Learning and Deep Learning

Abstract

The research titled “Feature Engineering for Students Academic Performance and Reignition Using Machine Learning and Deep Learning” highlights the transformative potential of modern computational methods in strengthening educational predictive analytics. Utilizing 3028 response collected via a through Google Form survey with 46 queries , the study digs into factors affecting student success, with feature engineering translating raw data into relevant feature for enhanced model accuracy. Among machine learning models Random Forest achieved the best accuracy of 94.61% and the lowest error rates, followed closely by Decision Tree with 93.785 accuracy. Gradient Boost , KNN, AdaBoost had accuracies of 92.29% 88.65% and 81.74% respectively, with AdaBoost performing the lowest. In deep learning models ,ANN stood out with a 94.50% accuracy , while the hybrid CNN-LSTM model followed with 93.48% accuracy. LSTM and CNN scored 93.27% and 93.195 accuracies, respectively with CNN performing the lowest among deep learning models. One essential part of this study was determining the most influential characteristics affecting student performance , with favorably and adversely. Features with correlation value greater than 0.4 or less than -0.4 were considered significant, including ; Gender, Department Choice, Study Hours Per Week, Family Responsibility, Past Failure, Average Sleep Hours, Study Year, Strategy of Measuring Performance_MCQ, Strategy of Measuring Performance of_Oral, Written, Assignment, Presentation. The goal of the research is to improve our knowledge of how feature engineering and deep learning can be used most effectively in this situation. This will help both the academic community and the real world of education, where better predictive models can directly and favorably affect student outcomes and the larger picture of educational success.