Stock market prediction with deep neural networks

OData support
Supervisor:
Dr. Bolgár Bence Márton
Department of Measurement and Information Systems

Stock market prediction refers to an activity that determines the future value of a company stock or other financial instrument. A model that can keep the rate of error within certain limits can be extremely useful in the success of this activity.

The development of the computing capacity opened new gates to the stock market prediction area. One of the most significant techniques is the use of Artificial Neural Networks as a mathematical function approximator. These networks are trained by the backpropagation algoritm. After the training they can generate outputs that are best suited to the learned inputs.

Kaggle is a platform that hosts different predictive modeling and analysis competitions with the help of statistical data from different companies. Since 2010, it has became one of the largest existing public community in data science.

The purpose of my work is to achieve the best possible rankings on this site by selecting the Winton company's listed challenge. The challenge is over, but the points and places of the users are still available. Additionally, the site is still evaluating the predictions I upload. The competition was issued with 832 valid predictions. The effectiveness of predictions is measured by the weighted average absolute error. The winner score is 1727, 53, while the last' s score is more than 30,000,000. The field is extremely tight, as the 780th place has a score of 1798.55.

Downloads

Please sign in to download the files of this thesis.