a hybrid feature selection and classification framework for predicting entrepreneurial competency using machine learning and binary grey wolf optimizer

Entrepreneurial competency is a fundamental element of economic development and innovation in education. research has shown that there is a relationship between psychological, educational, and environmental factors that lead to entrepreneurial capability. however, the combination of machine learning with accurate feature selection has not been thoroughly investigated. this study offers a hybridized framework combining the binary grey wolf optimizer and machine learning classification to predict the entrepreneurial competency of university students. the binary grey wolf optimizer features flexible options while grounded in the entrecomp framework and systematically captures multidimensional aspects of entrepreneurial competence as reflected in a theoretical framework. this work applied mutual information filtering, adasyn to resolve class imbalance, and optuna to adjust the hyperparameters for 16 different machine learning classifiers on a student data set with 219 records. the lightbgm algorithm demonstrated the highest level of accuracy, achieving 70.7%, followed by f1 with 68.3%, and an roc auc measure of 70.5%. furthermore, shap analysis clearly demonstrated that entrepreneurial environments, physical health, and resilience play a pivotal role in assessing an individual's entrepreneurial potential. a comparison with sharma and manchanda's (2020) benchmark, which utilized standard classifiers and achieved a maximum accuracy of 59.18%, showcases the clear benefits of this integrated, optimization-based method. in addition to offering greater accuracy, this study presents a scalable and interpretable framework for academic institutions to assess and encourage entrepreneurial growth and advancement. future research may expand upon this work by conducting longitudinal studies, utilizing cross-cultural datasets, and exploring alternative metaheuristic algorithms.