StockTrAIder: an ML-integrated stock trading platform

This work integrates multiple statistical and ML-based models within an accessible web application to analyze and predict a stock's future value for trading and market analysis. The application specializes in presenting key stock data, including historical patterns, long-term projections, and short-term portfolio actions. Dynamic Time Warping (DTW) models help identify similar time sequences to the current sequence and reveal the historical outcomes that preceded it. Statistical models, such as LightGBM, Random Forests, and AutoARIMA, use time-series and tree-based methods to make long-term predictions for next week's prices. A bagged XGBoostRegression model takes stock indicators as input to estimate the next day's closing price, providing guidance for specific portfolio allocations with time and market awareness. Portfolio holdings are defined by the Dynamic Strategy Selector, which evaluates previous portfolio performance and upcoming closing price predictions for all current S&P 500 companies. Predictions for portfolio-based actions are backtested from 2019-2025 to verify returns across a complete historical dataset, yielding an average annual return of 18.37%, an average maximum drawdown of -4.46%, and an average Sharpe ratio of 2.45. All model output is presented in an accessible, open-source dashboard for advanced and naive traders to understand current stock market conditions. 

Major Professor: Simanta Mitra