Mining Hidden Patterns in Student Dropout: Using Random Forest-Based Feature Selection and Association Rule Mining

Traditional dropout prediction models primarily focus on forecasting whether a student will leave, but often fail to uncover the underlying factors driving this outcome. To bridge this gap, this study utilizes the Apriori algorithm, which not only predicts dropout, but also reveals hidden patterns and relationships between influencing factors. We categorized the dataset into four groups: academic, socioeconomic, demographic, and macroeconomic, then performed association rule analysis with a minimum support of 5% and confidence of 50%. The initial findings revealed that academic performance had the strongest link to dropout. To observe the consistency and strength of the generated association rules, we gradually increased the confidence threshold from 50% to 100%, and only academic-related attributes continued to generate high-confidence rules. To further reinforce these results, we implemented Random Forest-based feature selection, identifying key features and applied the Apriori algorithm. This produced thirty rules with 100% confidence, 5% support, and a lift of 2.55, outperforming prior studies that reported 78% and 95% confidence with lift values of 1.02 and 2.0, respectively. Our model demonstrated a 22% and 5% higher confidence, along with an improvement in lift values over these studies.