Machine Learning Decision Support for Predicting and Profiling Student Startup Readiness in Higher Education Institutions

Abstract

This study develops a machine learning decision support framework for assessing startup readiness among higher education institution students in Pangasinan, Philippines. Survey data from 385 students were transformed into construct level indicators for startup knowledge, policy awareness, career aspiration, entrepreneurial skills, and business resource readiness. The analysis first examined reliability, associations, predictive relationships, and construct differences using Cronbach alpha, Spearman correlation, multiple regression, and the Friedman test. It then trained Logistic Regression, Decision Tree, and Random Forest models to classify practical startup readiness using academic profile variables, startup knowledge, and policy awareness as non-leakage predictors. K-Means clustering was applied to identify student readiness profiles for institutional decision support. The constructs showed acceptable to good reliability. Startup knowledge and policy awareness were stronger than practical readiness dimensions, confirming a relative knowledge action gap. Logistic Regression obtained the best classification result with 0.66 accuracy, 0.66 weighted F1 score, and 0.729 receiver operating characteristic area under the curve. Random Forest feature importance ranked policy awareness and startup knowledge as the strongest predictors, while clustering produced four intervention-oriented profiles. The study contributes a compact analytics workflow that higher education institutions can use to screen readiness, identify student groups, and design targeted entrepreneurship support.