Accepted at WI 2026 · Business Informatics · Student Track
A Data-Driven Decision Support System for Venture Capital Valuation
ESCP Business School · Paris / Turin · 2025–2026
Co-authored a conference paper based on my master’s thesis: an end-to-end ML pipeline for private-company valuation, combining investor co-investment networks, funding history, and market signals.
- PitchBook-backed deals
- 3,403
- Random Forest R²
- 0.557
- Mean absolute error
- 0.888
Key finding
Investor-syndicate capacity (mean co-investor AUM) was the dominant valuation driver in the SHAP analysis, providing evidence of information saturation.
Methods & evaluation
Engineered network features including centrality, clustering, and syndicate overlap. Benchmarked Random Forest and XGBoost against DCF/comparables, using SHAP for explainability and an out-of-time holdout spanning the 2022 correction. The best model performed competitively with, and more robustly than, the traditional baselines.
