Stock Price Forecasting

Self-defined course project for EECE 571T (Advanced Machine Learning Tools for Engineers)

Authors

Menghong Huang and Zhaosheng Li

Table of contents

• Description
• ML models
• Data in this project
• Forecasts
• Evaluation metrics
• Files in this repo
• Results
• Dependencies for this project
• Download the code

Description

This project implements various machine learning (ML) models to forecast the stock price of an American gaming merchandise retailer, namely, GameStop Corporation (GME). There are two types of forecasts in this project, which are the next-day forecast and the 30-day forecast. The objective of this project is to compare the performance of each model on those two kinds of forecasts, and the comparison is based on several evaluation metrics.

ML models

• Linear regression (LR)
• Support vector regression (SVR)
• Backpropagation neural network (BPNN)
• Simple recurrent neural network (RNN)
• Long short-term memory (LSTM) network
• Gated recurrent units (GRU) network

Data in this project

The dataset in this project, GME Historical Stock Prices, is publicly available on Kaggle. It is also generously offered by Yahoo finance website. The dataset records the daily data of GME stock exchange, and each entry has six features related to the stock price. The six features are open price, high price, low price, volume, close price, and adjusted close price

Forecasts

Next-day forecast

Most of the ML models in this project have the stock prices in the latest consecutive four days as input, whereas the LR and SVR models use the close price on today as input. The GME close price right after today is the target variable.

 

30-day forecast

Most of the ML models in this project have the stock prices in the latest thirty days in a row as input, whereas the LR and SVR models use the close price on today as input. The GME close price on the 30th day after today is the target variable.

Evaluation metrics

• Root mean square error (RMSE)
• Coefficient of determination (R²)
• Trend prediction accuracy (Accuracy)

Files in this repo

This GitHub repo contains the following folders or files

• *_model folders contain all the trained models used in this project

• data folder contains the dataset used in this project

• doc folder contains the documentation that is relevant to this work, which include a project report that has specific technical details and in-depth discussion

• img folder mainly contains the images of the training results, model structures, and test results

• BPNN_1.ipynb is the code for using BPNN to do next-day forecast

• BPNN_30.ipynb is the code for using BPNN to do 30-day forecast

• GRU_1.ipynb is the code for using GRU to do next-day forecast

• GRU_30.ipynb is the code for using GRU to do 30-day forecast

• LSTM_1.ipynb is the code for using LSTM to do next-day forecast

• LSTM_30.ipynb is the code for using LSTM to do 30-day forecast

• SVR2.0_1.ipynb is the code for using SVR and LR models to do next-day forecast

• SVR2.0_30.ipynb is the code for using SVR and LR models to do 30-day forecast

• simpleRNN_1.ipynb is the code for using simple RNN to do next-day forecast

• simpleRNN_30.ipynb is the code for using simple RNN to do 30-day forecast

• SVR.ipynb is out initial attempt on this project, which aims to conduct day-to-day forecast. But the result is far below our expectation and turns our prediction into the next-day forecast and 30-day forecast instead

Results

Next-day forecasting result

	LR	SVR	BPNN	Simple RNN	LSTM	GRU
RMSE	0.502	0.528	0.557	0.669	0.583	0.663
R2	0.993	0.993	0.992	0.988	0.991	0.988
Accuracy	50.1%	48.6%	47.6%	50.9%	49.6%	49.8%

LR	SVR

BPNN	Simple RNN

LSTM	GRU

 

30-day forecasting result

	LR	SVR	BPNN	Simple RNN	LSTM	GRU
RMSE	2.365	2.809	3.539	3.509	2.407	2.438
R2	0.848	0.786	0.654	0.660	0.841	0.836
Accuracy	51.3%	48.3%	51.1%	49.1%	48.9%	50.3%

LR	SVR

BPNN	Simple RNN

LSTM	GRU

Dependencies for this project

This project requires the following python modules:

numpy  matplotlib.pyplot  pandas  sklearn  keras  seaborn  joblib

Please make sure you have all the modules installed before running the code. For installing these modules, one can use command pip install or conda install

Download the code

git clone https://github.com/zhaoshengEE/Stock_Price_Forecasting.git