Comparing Machine Learning Models for Stock Market Price Prediction

CS 5805: Machine Learning Final Project

Pranesh Ambokar, Mohammad Heydari, Roshan Ravindran, and Leo St. Amour

Overview

This project analyzes and compares the performance characteristics and prediction accuracy trade-offs of linear regression and long short-term memory (LSTM) when applied to stock market prediction. We train the models using historical stock data and ask them to predict a stock's closing price given its opening, low, and high prices.

Structure

• Code: all of the code can be found in project.ipynb.
• Data: the data is located in the large_cap and mid_cap directories. The directories contain 50 large-cap and 50 mid-cap stocks, respectively.
• Saved data: the trained LSTM model is saved as lstm_trained.keras. The kt_dir directory contains cache files for LSTM hyper-parameter fine-tuning. The graphs directory contains saved versions of accuracy and day-trading simulation graphs.

Running the Code

To run the code, install any necessary dependencies:

$ python3 -m pip install numpy pandas matplotlib scikit-learn tensorflow keras_tuner

Then open the project using Jupyter, or a comparable system, and execute each cell in order.

NOTE: the final two cells, which execute the day trading simulation for linear regression and LSTM, take approximately 10 minutes and 1 hour, respectively, to execute.