A hybrid AI-based stock market prediction system using LSTM, Random Forest, and XGBoost, built for real-world deployment with Optuna-powered tuning, feature-rich engineering, and ensemble prediction logic. Designed to optimize F1 score and accuracy, this system aims to generate reliable buy/sell signals on stocks. still work under progress

This project integrates:

• 🔁 Recurrent Neural Networks (LSTM) for sequential financial patterns
• 🌲 Random Forest for ensemble-based classification
• ⚡ XGBoost for gradient boosting decision trees
• 🎯 Optuna for automatic hyperparameter tuning (optional mode)
• 📊 Backtesting Module to simulate trading performance

⚙️ Built by a Computer Engineering student to demonstrate real-world ML/AI skills in finance and time series prediction.

• ✔️ Ensemble of 3 models: LSTM + RF + XGBoost
• ✔️ Flag-based retraining (no need to retrain every time)
• ✔️ Real stock data from Yahoo Finance
• ✔️ Feature-rich engineering: RSI, Moving Averages, Volatility, Volume
• ✔️ Backtesting for historical performance validation
• ✔️ Soft voting for final trade signal (BUY / SELL)
• ✔️ CLI-based output, no GUI bloat
• ✔️ Saved model reuse (models/ folder)

LSTM-RandomForest-XGBoost-Stock-Predictor/ ├── LSTM+Random Forest+XGboost Stock Predictor.py ├── LSTM+Random Forest+XGboost Stock Predictor with optuna.py ├── backtest.py ├── models/ # Contains saved models ├── README.md ├── LICENSE # MIT License

pip install -r requirements.txt If no requirements.txt, install manually pip install yfinance numpy pandas scikit-learn xgboost tensorflow optuna

python "LSTM+Random Forest+XGboost Stock Predictor.py"

python "LSTM+Random Forest+XGboost Stock Predictor with optuna.py"

python backtest.py This will simulate trades and give statistics based on past 3 years of stock data.

📈 Processing RELIANCE... ✅ Trained RF | Accuracy: 0.72 | F1: 0.75
✅ Trained XGB | Accuracy: 0.74 | F1: 0.77
✅ Trained LSTM | Accuracy: 0.76 | F1: 0.79
🔎 RF Signal: BUY
🔎 XGB Signal: BUY
🔎 LSTM Signal: SELL
🟢 Final Signal: BUY

✅ Real-world applicability
✅ Combines traditional ML and DL
✅ Good F1/accuracy across multiple models
✅ Shows model saving, loading, and production-like behavior
✅ Demonstrates ability to tune, backtest, and validate system performance

This project is licensed under the MIT License.

Aarav Vinayak Mehta B.Tech Integrated Computer Engineering
MPSTME, NMIMS Mumbai Campus

If you're a company looking for a student with real-world skills in AI, ML, and applied engineering — this is just the start. Let's connect on LinkedIn or drop me a message!