SWIMMING RESEARCH

Design of Performance Predictor Models for Sprint Freestyle Swimmers Using Machine Learning Techniques

by
Libor Janek
and
César M. Guerra-Salcedo
and
Joaquín Pérez-Ortega
and
Rodolfo A. Pazos-Rangel

September, 2004

Abstract
This research addresses the problem of variable selection needed for consideration in a potential swimmer selection. The question that is posed could be stated as follows, ‘Is there a way to correctly predict the success of a sprint freestyle swimmer based on the subset of variables extracted by the feature selection method from a large dataset?’. This work describes the use of a genetic algorithm in combination with classification techniques such as decision trees and support vector machines in the process of selecting the most viable feature subset needed for maximized and accurate classification of swimmers. Results of the different classifiers were compared against each other. The research successfully presented the advantage of feature selection by increasing the correct classification rate of a learning model by as much as 25.37% and showed the need for an implementation of a sophisticated search algorithm to find the best suitable attributes for classification.

Keywords{Swimming Performance, Machine Learning, Model Selection, Data Mining}