Data-Driven Early Warning System for Subject Performance: A SMOTE and Ensemble Approach (SMOTE-RFET)
R. Hari Krishna, Pranav Vallabhaneni, Rudraraju Sri Krishna Chaitanya, Kiran Kumar Kaveti, M V A L Narasimha Rao, N S Koti Mani Kumar Tirumanadham
Abstract
Any country's economy must be strengthened in order to satisfy its citizens' healthcare and education requirements. The expanding data availability in Educational Data Mining (EDM) offers the opportunity for performing thorough investigation. Early academic success prediction is extremely valuable for detecting kids who are initially struggling, especially in subjects like Mathematics-1. The goal of this work is to forecast Mathematics-1 test scores using machine learning techniques, allowing for prompt intervention. Students' performance in related classes and their likelihood of passing Mathematics 1 on their first try are important determinants of prognosis. In order to determine the most effective method for improved prediction accuracy, the study examines a variety of machine learning algorithms, including Decision Trees (DT), Random Forest (RF), Logistic Regression (LR), K-Nearest Neighbours (KNN), and Naive Bayes (NB). This top-performing algorithm is then used in a bagging technique to increase forecast accuracy even more. The effectiveness of decision trees in forecasting student achievement is highlighted by prior research. Machine learning approaches, particularly classification algorithms, have been used in numerous researches to improve the prediction of student outcomes by utilizing a variety of strategies. The study makes use of a sizable dataset amassed from surveys of B.Tech students, encompassing characteristics like gender, success on the first try in Mathematics-1, and achievement in important disciplines. The suggested ensemble strategy, utilizing the Random Forest algorithm, outperforms individual algorithms in experiments by utilizing data preparation techniques like feature selection and SMOTE. It demonstrates increased specificity, sensitivity, precision, and accuracy in predicting student performance, which could help educators and institutions spot at-risk students and facilitate quick interventions. This study demonstrates the effectiveness of a Random Forest ensemble approach (SMOTE-RFEA) in improving the accuracy of predicting student achievement, underscoring the power of predictive analytics in education. Future research projects might focus on strategies that can be optimized to increase prediction accuracy and efficiency.