Predicts whether an IPO will be profitable at listing using subscription data.
The final model focuses on key investor demand signals (QIB and RII) and is optimized for high-precision decision-making.
- Predicts probability of IPO profitability
- Compared multiple models (Logistic Regression, Random Forest, XGBoost, Neural Networks)
- Performed feature engineering using subscription ratios
- Applied L1 regularization for feature selection
- Optimized decision threshold for high precision (~0.93)
- Deployed as an interactive Streamlit app
- Python, Pandas, NumPy
- Scikit-learn
- Streamlit
- Institutional subscription (QIB) is the strongest predictor
- Retail subscription (RII) provides additional signal
- Many features (price, size, HNI, ratios) were removed via L1 regularization
- Final Features: Subscription_QIB, Subscription_RII
- ROC-AUC: ~0.75
- Precision: ~0.93 (at threshold = 0.55)
- Recall: ~0.43
pip install -r requirements.txt
streamlit run app.pyTraining notebook: Colab file