Prediction of Sporting Results using Neural networks or Machine Learning Techniques

Student: Michael Stanley James Baulch

Supervisor: Dr. Marcus Gallagher

Category: Computer Systems Engineering Thesis Project

Artificial neural networks, or ANN’s as they are commonly called, are used in many places throughout the world for a wide variety of applications. One of these applications is to predict results of sporting matches, and this is the section that I will be focusing my study on. I will be concentrating directly on rugby league and basketball. Training an ANN can be a long and tedious task, however with a bit of persistence and a lot of patience, accurate results can be achieved. By looking at the input and output data that has been entered by a user, the neural network looks at possible solutions as to how the output was achieved from the input. The ANN then tries to map these solutions to the other data points that have been entered, and checks to see how accurate it actually was. A typical artificial neural network has three layers of neurons: an input layer (equivalent to biological senses), a "hidden" layer that does the processing, and an output layer that provides the answer. There are many different methods of training an ANN, however I will be focusing mainly on the Back Propagation method. A data set has already been decided upon, and consists of 20 inputs and 2 outputs (only 18 inputs for basketball). The network is currently being trained and improved each day. Different methods are being tested to try and obtain a more accurate result. For example, instead of using back propagation, I will try the Quasi-Newton method, conjugate gradient, and Levenberg Marquardt and see if the performance is improved. Other properties of the network can also be changed, such as the number of ‘hidden’ layers in the network, and so on. While accurate results cannot always be guaranteed, hopefully in this project, by taking care as to what data and methods to use when training the network, good results will be achieved.

Poster Presentation (PDF)

Thesis Document (PDF)

Other Related Files