Proof of Concept / Gradient Boosting Regression
Gradient Boosting on Historical Oncology Launch Analogues
A gradient boosting regression model trained on 20 synthetic oncology launch analogues to predict a new product's market share trajectory over 24 months. Features include clinical evidence strength, market structure, access barriers, and commercial execution — with bootstrap confidence intervals and analogue-based benchmarking. Built entirely in-browser.
Disclaimer: This is a proof-of-concept using synthetic launch data and fictional product names. All market share projections, launch characteristics, and analogue profiles are illustrative. This tool is NOT intended for actual commercial forecasting or investment decisions.
Generating historical analogues & training gradient boosting model...
Training data generated from 20 synthetic oncology launches using logistic growth curves with feature-dependent parameters. Each analogue contributes 24 monthly observations, yielding 480 training samples across 11 feature dimensions (10 launch features + month number).
10 gradient boosting models trained on resampled analogue sets. For each prediction month, the 2.5th and 97.5th percentile of ensemble predictions define the 95% confidence band, capturing model uncertainty due to limited training data.
Feature vectors min-max normalized across all analogues, then compared via cosine similarity. Top matches serve as intuitive benchmarks — showing the user which historical launches most closely resemble their configured scenario.