ayushdayal8/IPO-Success-Predictor

Advanced IPO Success Predictor Report

This report describes the implementation of an Advanced IPO Success Predictor using machine learning with an emphasis on sophisticated feature engineering, ensemble modeling, and a dynamic prediction system integrated into a Streamlit web application.

Overview and Purpose

The system predicts IPO outcomes classified into three categories:

• —Success (S)
• —Normal (N)
• —Fail (F)

The model leverages financial metrics and market sentiment to classify IPOs with high accuracy, facilitating investors' decisions.

Data Loading and Preprocessing

# Replace NA values and handle missing data df = df.replace(['NA', '', 'NaN', 'null'], 0) df = df.fillna(0)

return df except Exception as e: st.error(f"❌ Error loading dataset: {str(e)}") return None

• —Dataset is loaded and cleaned.
• —Missing or malformed values are replaced with zeros.
• —Data caching improves app responsiveness.

Advanced Feature Engineering

• —Extracts critical financial features, including P/E ratios, ROCE, market capitalization, dividend yield, profits, and sales.
• —Creates interaction and polynomial features to represent growth, stability, and valuation risks.
• —Encodes target labels for classification.

Notable engineered features include:

• —PE_ROCE_Interaction
• —Profit_Margin
• —Market_Cap_to_Sales
• —Profit_Growth_Momentum
• —Valuation_Risk
• —Efficiency_Score

This enriches the feature space to capture complex relationships.

Ensemble Model Construction

The model ensemble includes:

• —Base models are combined using a Stacking Classifier with Logistic Regression as meta-learner.
• —A Voting Classifier provides a backup ensemble strategy.

Training Pipeline

• —Features are scaled using the PowerTransformer (Yeo-Johnson).
• —Recursive feature elimination with cross-validation (RFECV) selects the most informative features.
• —Class imbalance is addressed with the SMOTETomek approach (oversampling + cleaning).
• —Stratified train-test split ensures balanced classes.
• —Individual and ensemble models are trained and evaluated with key metrics: accuracy, F1-score, precision, recall.

Prediction System with GMP Integration

• —Users input IPO details plus optional Grey Market Premium (GMP).
• —The system performs feature engineering and scales input features.
• —GMP is incorporated dynamically with a weight dependent on its ratio value, adjusting predictions.
• —Confidence calibration considers probability distributions to adjust classification thresholds.
• —Outputs include final class prediction, adjusted scores, confidence percentages, and probability breakdowns.

Streamlit Web Application

• —Interactive interface supports three modes:

1. 1.Model Training: Load data, engineer features, train ensembles, and save models.
2. 2.IPO Prediction: Input new IPO attributes to receive success predictions and investment recommendations.
3. 3.Model Analysis: Visualize performance metrics and confusion matrices for model evaluation.

• —Visualizations include:

• —Bar charts comparing model accuracies and F1-scores.
• —Confusion matrix heatmaps.
• —Color-coded performance scales indicating risk and opportunity.

Conclusion

This solution integrates:

• —Domain knowledge through engineered financial features.
• —Advanced ensemble learning techniques.
• —Dynamic market sentiment incorporation with GMP.
• —Robust evaluation with clear visual and textual outputs.

It empowers investors with actionable IPO success predictions supported by rigorous data science and machine learning methodologies.

Appendix: Key Functions

• —load_and_preprocess_data: Load and clean dataset.
• —advanced_feature_engineering: Generate enriched features.
• —create_advanced_ensemble: Build stacking and voting classifiers.
• —train_advanced_model: Perform feature selection, balancing, training, and saving.
• —predict_with_gmp_advanced: Predict IPO success with GMP-based score adjustment.
• —display_model_performance: Visualize and summarize model performances.
• —Streamlit UI functions: main, show_model_training, show_ipo_prediction, show_model_analysis.

This concludes the report on the Advanced IPO Success Predictor.