A simulation study on clustering time series with metaheuristic methods
Roberto Baragona · IRIS Research product catalog (Sapienza University of Rome) · 2001
Given a set of time series, let be the cross-correlation matrix function which may be computed from prewhitened residual series. Then, a dissimilarity index is assumed between each pair of time series which accounts for the crosscorrelations but does not necessarily fulfills the Euclidean distance requirements. Metaheuristic methods are proposed to partition the set of time series into clusters in such a way that ( ) the cross-correlation maximum absolute value between each pair of time series that belong to the same cluster is greater than some given threshold, and ( ) the -min cluster criterion is minimized. The simulation experiment shows that suitably designed metaheuristic methods, and especially the tabu search algorithm, are able to solve this problem successfully, and produce results better than both the single linkage method, and, on the other hand, a pure random search algorithm.