Impact
Built an investment analysis tool that combined value investing principles with machine learning, enabling more reliable stock buy/sell decisions; achieved forecasting accuracy with LSTM (MAE 0.02, RMSE 0.03) and demonstrated intrinsic valuation on $1T+ market cap companies (Apple, Microsoft)
Problem Statement
Traditional retail investors lack the time and expertise to assess whether a stock is undervalued or overvalued. Existing approaches often rely on short-term signals, ignoring fundamentals, and offer limited predictive power for long-term decision-making.
Approach
The development strategy focused on a phased implementation:
- Integrate financial fundamentals with machine learning forecasting
- Provide investors with a simple webapp that outputs whether a stock is fairly priced and predicts its near-term trend
- Blend classical valuation (DCF + Monte Carlo Simulation) with modern ML (LSTM, Random Forest)
Methodology
Data Collection
- Pulled 5 years of financial and stock market data using Alpha Vantage and Yahoo Finance APIs.
- Extracted key financial ratios (P/E, P/B, Free Cash Flow, D/E, PEG)
Valuation
- Applied Discounted Cash Flow (DCF) model
- Ran Monte Carlo simulations to calculate intrinsic value distributions.
Forecasting
- Implemented stacked LSTM in PyTorch for 6-month price prediction.
- Benchmarked against Random Forest and logistic regression.
- Evaluated using MAE, RMSE, Accuracy, Precision, Recall, F1.
Deployment
- Built a Flask backend + Angular frontend
- Delivered intuitive buy/sell signals and dashboards to end users
Result
- Intrinsic Value Analysis: Correctly flagged Apple and Microsoft as overvalued during evaluation period.
- Forecasting: LSTM — MAE 0.0239, RMSE 0.0317 (best performer); Random Forest: Less accurate, but interpretable baseline.
- End-User Impact: Provided actionable buy/sell recommendations in a webapp, simplifying complex financial analysis for non-technical investors.