Supervised Stock Price Prediction using Long-Short Term Memory and Technical Indicators
This dissertation focused on building LSTM models that can predict stock price values and trends any number of days in the future of any stock index, public company, or cryptocurrency using historical time series of 52 technical indicators. The models from the dissertation had achieved promising results by beating persistence baseline models and by outperforming 6 out of 7 selected research papers in the same domain. The best models for the companies in Nasdaq 100 achieved 61.8% Trend Correct Percentage (TCP) and 0.80% Mean Absolute Percentage Error (MAPE) for predicting 1 trading day, 58.2% TCP and 1.46% MAPE for predicting 3 trading days, 56.6% TCP and 1.82% MAPE for predicting 5 trading days, and 56.6% TCP and 2.59% MAPE for predicting 10 days in the future.
This repository contains the following:
- The dissertation report
- The final presentation
- Code developed for the models