International Journal of Business and Economics Volume 25, No. 1 June, 2026

Not Just Luck: Predicting Startup Success with Machine Learning

Shobhanam Krishna

Organisational Behaviour and Human Resources, Indian Institute of Management Shillong, India

Anita Choudhary

Organisational Behaviour and Human Resources, Indian Institute of Management Shillong, India

Rohit Dwivedi

Organisational Behaviour and Human Resources, Indian Institute of Management Shillong, India

Ashutosh Bishnu Murti

Organisational Behaviour and Human Resources, Indian Institute of Management Shillong, India

Abstract

Startups are central to innovation but face a staggering failure rate of nearly 90%, largely due to misaligned product-market fit, limited capital, flawed business models, or ineffective leadership. Traditional methods for predicting startup outcomes often fail to account for the complex, nonlinear dynamics of entrepreneurial ecosystems. This study aims to identify critical success factors and develop a predictive model to assess startup viability using supervised machine learning, thereby improving investment decision-making and supporting ecosystem resilience. Leveraging a global Crunchbase dataset with over 100 multidimensional features, the study employs Principal Component Analysis (PCA) for dimensionality reduction and Random Forest for classification. A five-step data mining framework was used to process, transform, and analyze the data. The model classifies startups as successful (acquired/operating) or failed (closed), with predictive accuracy evaluated through confusion matrices and F1-scores. The model achieved 93.9% test accuracy and 0.82 confusion matrix accuracy. Key predictors include founder background, funding patterns, industry, geography, and early-stage agility. Practical experience and consulting exposure were stronger predictors than formal education. PCA helped enhance interpretability and reduce noise. The study bridges Human Capital Theory with machine learning, offering a rare integration of theoretical grounding and algorithmic modeling. Unlike prior research, it combines PCA with Random Forest, enabling robust, interpretable predictions on a large-scale, global dataset. The findings inform investors, incubators, and policymakers on optimizing startup selection, funding strategies, and ecosystem development. The model serves as a scalable tool for enhancing startup evaluation, resilience, and long-term innovation impact.

Keywords：Crowdfunding Platforms, Data Mining, Machine Learning, Marketing Strategies, Startup Success

JEL Classifications：M12, M13, M14, M15, M51, M53, M54, O15, E24

BACK