SDSU Department of Computer Science
  
Home Contact Us Undergraduate Student Info Graduate Student Info Faculty & Staff News & Events
News Calendar Problem of the Fortnight

February 2009 Calendar

CS Masters' Thesis Defense

Title: Financial Prediction Through Use of a Genetic Algorithm
Speaker: Rachel Schneider
Date: Thursday, February 19, 2009
Time: 9:30 a.m.
Location: GMCS 405
Thesis advisor: Dr Joseph Lewis

Abstract:
The Financial Prediction Planner provides users the ability to predefine expense categories and enter financial expenditures into these categories. The application allows financial expenses to be tracked and displayed based upon such aspects as the date, purchase price, and category. This paper describes an experiment which incorporated a genetic algorithm into the Financial Prediction Planner to create an application which is capable of utilizing a user’s previous financial data in order to predict future financial expenses.

Populations were created from two types of agents – ones which predicted one expense category at a time and others which predicted four categories simultaneously. The agents were run through fitness tests in which they predicted future financial expenses based upon past financial data existing in the personal financial planner system. Roulette wheel selection, mutation, recombination, and migration were used with the genetic algorithm to run multiple sets of each type of population through many generations. At the conclusion of this process the most optimal agents from each type of population (single and multiple predictors) were used to predict the financial expense of a future month as well as the range of months used for the fitness tests. These predictions were then compared to the actual values. Fairly accurate predictions (all on average were within twenty-five percent of the actual value) were obtained by the agents over a range of months. It was concluded that the agents which predicted for multiple expense categories simultaneously were capable of producing more precise predictions on average over the agents which predicted for a single expense category at a time. This paper describes the methods of the study as well as the results and future recommendations for continuing experimental research.
webmaster@cs.sdsu.edu page counter College of Sciences