A Weighted Local Mean-Based k-Nearest Neighbors Classifier for Time Series

Abstract

We propose a Weighted Local Mean-based k Nearest Neighbors (WLMk-NN) classifier for time series. The proposed method is different from Local Mean-based k-Nearest Neighbors (LMk-NN) in that it assigns a weight for each element of time series when calculating local mean vectors. Indeed, our proposed method determines the local mean vectors more efficiently than LMk-NN does by using some weights which are obtained from the distances of a query instance to its k nearest instances in each class. Our experiments were conducted over 85 time series datasets in UCR Time Series Classification Archive and compared with some baseline methods. Experimental results show that our method outperforms k-NCN, LMk-NN, LMk-NCN, and the widely used classifier k-NN in many time series datasets.