Abstract—The extreme volatility of the global startup land- scape presents a significant risk for venture investors and ecosystem stakeholders. Accurately determining the trajectory of a nascent firmwhether it will transition into a successful market exit or terminate its operationsremains a complex predictive challenge. This research presents a specialized machine learning framework aimed at evaluating startup durability, utilizing a comprehensive repository of 923 startup profiles characterized by 15 diverse features across financial, operational, and geo- graphical dimensions. To uncover non-linear patterns within the venture capital data, we introduced several synthetic indicators, including capital efficiency metrics and time-series milestone tracking. We evaluated three primary learning architectures: Logistic Regression, Random Forests, and Gradient Boosting, and subsequently developed a soft-voting ensemble mechanism to unify their predictive power. Our results indicate that the tuned Random Forest model achieves a high classification accuracy of 81.3% and an F1-score of 0.865. The findings highlight that algorithmic feature engineering significantly improves the reliability of risk assessment in the startup domain. Additionally, we discuss the implementation of a cloud-based inference system for real-time startup valuation, offering a robust quantitative tool for financial decision-makers. Index Terms—Entrepreneurial Risk, Predictive Analytics, En- semble Learning, Startup Success, Venture Capital, Machine Learning Applications
Rangaswamy et al. (Tue,) studied this question.