Machine learning models for early‐stage investment decision making in startups

Abstract

This study demonstrates the efficacy of machine learning techniques for evidence‐based evaluation of early‐stage ventures. Leveraging real‐world data on 24,965 startups across diverse sectors and countries sourced from Crunchbase, we develop predictive models using algorithms including random forest, XGBoost, and support vector machines. Rigorous training and testing on a 70–30 split of the data reveal that the algorithms can effectively classify startups as successful or not, achieving over 90% accuracy. Random forest emerges as the top performer, followed closely by XGBoost. This research demonstrates the immense potential of machine learning techniques in forecasting startup success to inform management practice.