ABSTRACT
Data classification has been widely studied in engineering, market and financial analysis, medicine and biology, and other areas. In particular, there exists a large body of literature concerning classification algorithms; however, the performance of such algorithms is highly data dependent. For example, while the k-Nearest Neighbors algorithm (k-NN) is quite versatile in its application, most parametric classifiers are limited in their use by assumptions. Using data from Sportsvision’s PITCHf/x, we apply several generic classification methods to the problem of pitch classification. We place a particular emphasis on improving the accuracy of Bayesian classifiers through feature selection and dimension reduction via Linear Discriminant Analysis (LDA) and Principle Component Analysis (PCA), respectively. The accuracy and speed of these classification algorithms are then analyzed and compared.
