Predicting IPO Listing Outcomes in the Indian Market: A Machine Learning and Deep Learning Approach with Explainable Risk and Investment Scoring

Initial public offering (IPO) listing outcomes are of substantial interest to retail investors, underwriters, and regulators, yet remain notoriously difficult to predict from pre-listing fundamentals alone. This study develops and evaluates a complete predictive modelling workflow for the Indian IPO market, combining regression analysis and deep learning to estimate four core outputs: an interpretable IPO Risk Score (0–100), an IPO Investment Score (0–100), the probability of a successful (positive) listing-day return, and the expected listing-day gain percentage. From an assembled dataset of 880 Indian mainboard IPOs (2006–2026) compiled from Chittorgarh, SEBI, the National Stock Exchange (NSE), and yfinance, 825 verified, leakage-free offerings were finalized after systematically resolving target listing returns and sanitizing 48 active fundamental, valuation, and macroeconomic features. In classification, a calibrated HistGradientBoosting model (ROC-AUC: 0.617, Log Loss: 0.635, Balanced Accuracy: 57.73% at optimal threshold 0.670) and a 10-fold bagged regularized deep multilayer perceptron (ROC-AUC: 0.618, Log Loss: 0.756, Balanced Accuracy: 57.73%) were compared and ensembled. Blending both paradigms yielded the strongest overall discrimination (ROC-AUC: 0.626, Log Loss: 0.645, Balanced Accuracy: 58.18%) with a 70.9% inter-model prediction agreement. For listing-day gain regression, HistGradientBoosting with absolute-error loss achieved MAE = 25.49% and RMSE = 38.50% (R² = 0.0381), consistently outperforming the naive mean baseline (MAE: 27.58%, RMSE: 39.29%), whereas deep learning neural regressors (including a 3-year sequential LSTM) encountered an information ceiling on tabular returns. Model explainability via SHAP highlights short-term index momentum, trailing revenue, balance-sheet scale, and composite scorecards as primary drivers. Finally, the complete dual-engine architecture was operationalized into an interactive web application featuring real-time prospectus scraping for ongoing IPOs and live index tracking, providing an actionable, uncertainty-aware decision-support tool for retail investors.