Systematic Literature Review on Entrepreneurship Potential Prediction.

Abstract

This systematic literature review synthesizes 48 empirical studies on “Machine Learning Models for Predicting Entrepreneurial Potential” to address gaps in understanding predictive feature, methodological rigor, and contextual adaptability. The review evaluates traditional statistical methods and machine learning (ML) for predicting entrepreneurial potential in individuals. The research indicates that hybrid machine learning models integrating feature selection and metaheuristic optimization—such as GLLCSA-KELM-FS and WHO-RXGBoost—attain superior prediction accuracy (85–95%) compared to traditional methods like SEM and logistic regression (65–75%). Although demographic and environmental factors provide contextual explanatory power, psychological constructs—particularly self-efficacy, risk-taking propensity, and creativity—prove to be the most reliable predictors. Despite methodological advancements, significant limitations persist, including cultural bias in datasets (73% of research focused on Asian contexts), challenges in model interpretability, and ethical concerns around algorithmic fairness. The paper identifies four primary topics for additional investigation: the development of explainable artificial intelligence frameworks; cross-cultural validation studies; dynamic evaluation systems for longitudinal prediction; and the implementation of justice-aware algorithms.

These findings have significant implications for policy and entrepreneurial education, suggesting that next-generation prediction systems must balance scientific complexity with practical usefulness. By addressing current constraints in generalizability and ethical implementation, the sector may evolve towards more equitable and practical models that effectively support emerging entrepreneurs across diverse socioeconomic contexts.

DOI: https://doi.org/10.5281/zenodo.21135607