Enhancing Graduate Employability Prediction Through Ensemble Learning For Educational Decision Support

Abstract

In higher education, employability prediction has become an important research domain wherein, universities increasingly employ methods to predict employability and their associated patterns. In this context, ensemble models tend to outshine traditional single models inherently due to their high prediction accuracy, low variance and reduced bias. Also, as per the survey of India Skill Report 2025 in India, Punjab (state) and Chandigarh (UT) have shown lowest grade of employability pattern. Therefore, for the purpose of this study, data is collected through a questionnaire from colleges and universities across Chandigarh and Punjab in India to study the pattern and prediction of employability. The Cronbach's Alpha value for the questionnaire was 0.947 which indicates reliability of the data collected. The proposed methodology uses four machine learning models—Support Vector Machines, Multilayer Perceptron, Random Forest, and K-Nearest Neighbors to predict students' employability. In this study, the ensemble approach provides a strong predictive framework that uses genetic algorithm to optimize learning parameters and improve adaptability. The use of genetic algorithm helps in identifying and discarding suboptimal candidate solutions which further helps in building iterative candidate solution population that shows efficient convergence. The performance of the proposed method is computed using accuracy, precision, recall, F1-score, support, and ROC–AUC score. The proposed model provides 91.8% prediction accuracy on the primary dataset and 85.5% accuracy on an external Kaggle dataset. The findings highlight that the model offers high predictive capability and generalizability for graduate employability prediction.