Data-Driven Revolution in Academic Support for Mathematics Underachievers through Random Forest Individual and Hybrid Model

In today's educational landscape, an abundance of student data is stored within institutions, yet its practical application to address academic challenges, like poor performance in specific courses, remains constrained. Educational Data Mining, leveraging machine learning and data mining techniques, aims to predict student performance using available datasets. This study specifically targets the improvement of underachievement in mathematics by analyzing initial and final grades. The dataset utilized in this study encompasses 30 distinct parameters, organized into four comprehensive categories: demographics, family details, study habits, and lifestyle choices. Efforts to enhance student performance and aid educators involve creating systems that scrutinize various factors ranging from prior assignment scores to social influences, parental occupation, and absenteeism to forecast academic success. Here, the Random Forest Classifier, a soft computing tool, was utilized on a dataset encompassing 649 student records. Furthermore, metaheuristic algorithms like Smell Agent Optimization and Giant Trevally Optimizer were employed to optimize model's hyperparameters, with the intention of enhancing accuracy and precision in performance estimations. The study reveals that the RFGT hybrid model resulting from enhancing the RFC with GTO outperforms other models, boasting notably high accuracy values of 0.947 and 0.958 for G1 and G3, respectively. This hybrid model's effectiveness is underscored by metrics that surpass those of alternative approaches significantly. The robust performance of the RFGT model holds promise in revolutionizing academic guidance and support for students, potentially enabling institutions to provide tailored interventions for struggling students early in their educational journey, thereby fostering improved outcomes for all.