Generic placeholder image

About the Project

Deep learning with its contributions to Artificial Intelligence has drawn researchers and investors to utilize it to predict stock price movement. In this project, we employ traditional machine learning models including linear regression, logistic regression and SVM as well as deep learning models including MLP and 1D CNN to predict the value and trend of Dow Jones Industrial Average. .

Generic placeholder image

About Us

Supervisor: Kwok-Ping CHAN

Students: LIU Jiayao(3035142596)

Generic placeholder image

Documents

The documents about ths project. Click buttons below to view more

Project Plan »

Interim Report »

Final Report »


Schedule & Progress

Here is our history...

Date Events
Oct 1st Deliverables of Phase 1: Inception
- Complete the project plan
- Build a project website
Sept 1st to Oct 15th Study the theory & related knowledge
- Deep learning
- Python programming
- Financial products
Oct 16th to Jan 10th Collect & preprocess the data
- Determine a predicted objective
- Collect data from WSJ and Yahoo Finance
- Clean and label the data
Jan 12th First Presentation
Jan 21th Deliverables of Phase 2: Elaboration
- Conduct preliminary implementation
- Complete detailed interim report
Jan 10th to Feb 25th Build the models based on the data
- Fit the model
- Tune the corresponding hyperparameters
March to April Finalize & Improve the results:
- Optimize the model
- Prepare for final presentation
- Start to write final report
Apr 15th Deliverables of Phase 3: Construction
- Conduct finalized tested implementation
- Complete the final report
Apr 17th Final presentation
May 2nd Project exhibition

Methodologies & Results

In this project, given the past stock data of Dow Jones Industrial Average (DJIA), we apply regression models to predict the value of DJIA and classification models to predict the trend of DJIA on the 31st day. The stock data of DJIA from Oct 5, 1987 to Sept 29, 2017 are collected from Yahoo Finance and split into training and test sets. The Mean Squared Error (MSE) of regression models as well as the accuracy of classification models on the test set are selected as the evaluation metrics. The model training results are as follows:

Regression Models

Model Test MSE
Linear Regression 18352.1540
MLP (regression) 18226.6978

Classification Models

Model Test Accuracy
Logistic Regression 0.5355
Support Vector Machine 0.5319
MLP (classification) 0.5487
1D CNN (univariate) 0.5548
1D CNN (multivariate) 0.5548

Name Email Phone
Kwok-Ping CHAN (supervisor) kpchan@cs.hku.hk
LIU Jiayao jadeliu@hku.hk (+852) 67649723