2011 Honors Theses
|Author: Brett Beaulieu-Jones
|Title: A fusion model of Artificial Neural Networks and Moore-Penrose Pseudo-inverses for Stock Market Forecasting|
|Advisor: Sergio Alvarez
This project proposes and implements a fusion model that combines a machine learning method, Artificial Neural Networks (ANN), with a linear algebra method, Moore-Penrose Pseudo-inverses, for stock market forecasting. The goal of the model is to decide which stocks to buy on a given day, by predicting the stocks expected to change in price by the greatest percentage. The prediction tool is trained with past daily closing prices, and tested with the previous four days' closing prices. It then uses a design by committee approach to combine the results of the ANN and the Moore-Penrose methods. Forecasts are evaluated with three experiments: a three month prediction on the dataset used by a leading academic project in stock market prediction, historical yearly predictions to test long term viability, and a five week live simulation experiment in an attempt to demonstrate real time practical use of the proposed method.