Sensor Data Mining on the Kinematical Characteristics of the Competitive Swimming

Abstract

The purpose of this study was to propose a new methodology for the automatic identification and the classification of the swimmers kinematical information during interval training of competitive swimming. Forty-five college swimmers attached the newly developed chest band sensor unit, which has a triple-axes accelerometer inside, and then performed a controlled interval training set with four stroke styles. The authors identified swimmer's states, such as the swimming/rest phases and the start, turn and goal touch events by using the trunk longitudinal acceleration (Ay). With the inductive inference based on the experimental results and the deductive inference based on the empirical rule on the interval training brought the estimation of the swimming time. For the classification of the swimming strokes, using the extracted swimming phase acceleration, the mean, variance and skewness of each bout were calculated. The authors compared different data mining algorithms for the stroke style classification with these descriptive statistics, such as mean, variance, skewness on the each axial acceleration as the independent variables and stroke styles as the depending variable. The accuracy of the stroke style classification by both the multi-layered neural network (NN) and the C4.5 decision tree were 91.1%.