Haar Wavelets for Online-Game Player Classification with Dynamic Time Warping

Online game players’ action sequences, while important to understand their behavior, usually contain noise and/or redundancy, making them unnecessarily long. To acquire briefer sequences representative of players’ features, we apply the Haar wavelet transform to action sequences and reconstruct them from selected wavelet coefficients. Results indicate that this approach is effective in classificationwhen the k-nearest neighbor classifier is used to classify players based on dynamic time warping distances between reconstructed sequences.