Paper Number
ICIS2026-1661
Paper Type
Completed
Abstract
Venture capital (VC) decision-making increasingly relies on data-driven approaches to reduce uncertainty in startup investments. However, while advanced machine learning models offer high predictive performance, their opacity limits transparency and trust. This study examines whether intrinsically interpretable machine learning (IIML) models, particularly advanced generalized additive models (GAMs), can achieve competitive performance while preserving interpretability. Following the knowledge discovery in databases process, we use data from 20,172 early-stage startups and compare advanced IIML models with traditional IIML models and black-box models in terms of predictive performance and functional interpretability. We further investigate the domain adaptability of advanced IIML models regarding interaction effects and monotonic constraints. Among other findings, our results show that IIML models achieve predictive performance comparable to black-box models while providing insights into key drivers. The study highlights the practical value of GAM-based approaches for VC decision support and contributes to research at the intersection of IIML and finance.
Recommended Citation
Zilker, Sandra; Käppel, Martin; Jobst, Hanna; and Weinzierl, Sven, "Leveraging Intrinsically Interpretable Machine Learning for Startup Success Prediction" (2026). ICIS 2026 Proceedings. 3.
https://aisel.aisnet.org/icis2026/innov/innov/3
Leveraging Intrinsically Interpretable Machine Learning for Startup Success Prediction
Venture capital (VC) decision-making increasingly relies on data-driven approaches to reduce uncertainty in startup investments. However, while advanced machine learning models offer high predictive performance, their opacity limits transparency and trust. This study examines whether intrinsically interpretable machine learning (IIML) models, particularly advanced generalized additive models (GAMs), can achieve competitive performance while preserving interpretability. Following the knowledge discovery in databases process, we use data from 20,172 early-stage startups and compare advanced IIML models with traditional IIML models and black-box models in terms of predictive performance and functional interpretability. We further investigate the domain adaptability of advanced IIML models regarding interaction effects and monotonic constraints. Among other findings, our results show that IIML models achieve predictive performance comparable to black-box models while providing insights into key drivers. The study highlights the practical value of GAM-based approaches for VC decision support and contributes to research at the intersection of IIML and finance.
When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.
Comments
15_Innovation_and_entrepreneurship