Portfolio Optimization Using Machine Learning

Developed a hybrid ML framework for dynamic portfolio optimization using supervised learning, time series forecasting, and reinforcement learning.

Key Features

• Created a custom OpenAI Gym environment to simulate multi-asset portfolio management
• Predicted asset returns using Random Forest and LSTM models
• Implemented Actor-Critic reinforcement learning to adjust portfolio weights dynamically
• Achieved improved Sharpe ratio (0.284) and better drawdown control compared to traditional strategies

Tools Used

Python, scikit-learn, TensorFlow, OpenAI Gym, Matplotlib, NumPy

Status

Project completed under academic mentorship as part of a research initiative.
Paper submission to iCWiCOM 2025 is planned (under review for Springer publication).

Future Work

• Add backtesting interface using Streamlit
• Tune RL agent for real-time data input