Fast and Accurate Temporal Data Classification Using Nearest Weighted Centroid

Classification is an important task in the field of machine learning and data mining due to its wide applications. Among many algorithms to classify temporal data, the k-nearest neighbor with the dynamic time warping (DTW) is performant. However, it is very time consuming as it needs to compare time series query with all training samples. Since the computational complexity of the DTW is quadratic, these comparisons pose some problems for large data, either it is not applicable or it takes a lot of time to classify the data. To obtain fast accurate results by reducing the data size, classification based on nearest centroid could be one solution. In this paper, to speed up the nearest neighbor classification, and specially to make it applicable for huge datasets, we propose a fast accurate k-nearest weighted centroid classifier. Using generalized $k$ -means-based clustering within each class, we obtain a small number of weighted centroids per class. A wide range of datasets is used to evaluate the efficiency of the proposed classifier, which outperforms the alternative ones.