Comparative Analysis of Machine Learning and Deep Learning Models for Predicting Student Selection Outcomes Using Examination Data
Main Article Content
This study presents a comparative analysis of machine learning and deep learning models for predicting student selection outcomes using examination data. The models evaluated include Logistic Regression, Random Forest, Extra Trees, Gradient Boosting, and a Multilayer Perceptron. The dataset consists of structured features such as examination scores and demographic attributes, with preprocessing steps including handling missing values, feature engineering, encoding categorical variables, and addressing class imbalance through class merging. Model performance was evaluated using accuracy, precision, recall, and F1-score, along with 5-fold cross-validation to assess robustness. The results show that Gradient Boosting achieved the best overall performance, followed closely by Random Forest and the Multilayer Perceptron, while Logistic Regression performed significantly worse. Feature importance analysis indicates that academic performance variables, particularly the total score and its components, are the most influential predictors, whereas demographic features have minimal impact. These findings suggest that ensemble-based methods are highly effective for structured tabular data and that deep learning does not necessarily outperform traditional machine learning models in this context. This study highlights the importance of selecting appropriate models based on data characteristics and provides insights for developing AI-based student selection systems.