Spatial Clustering Algorithm for Time Series Rainfall Data Using X-Means Data Splitting

摘要

The aim of this study is to present a new spatial clustering process for time series data. It has become an important and demanding application when the data involves chronological long time series and huge datasets. A great challenge in clustering is to achieve an optimal solution in searching similarity along the series. Furthermore, it also involves a very large-scale data analysis. Unfortunately, the existing clustering time series algorithms have become impractical since data do not scale properly for longer time series. The performance of the clustering algorithm gets even worse if it relies on actual data and many clustering algorithms are often faced with conflict in handling high dimensional data. In the case of spatial time series, the problem can be solved by unsupervised approaches rather than supervised classification, with appropriate preprocessing techniques to transform the actual data. The unsupervised solution using time series clustering algorithms is capable to extract valuable information and identify structure in complex and massive datasets as spatial time series. Therefore, a clustering algorithm by introducing data transformation using X-means data splitting is proposed to investigate the spatial homogeneity of time series rainfall data. The hierarchical clustering was used to demonstrate the similarity once the data was divided into training and testing sets. The proposed algorithm is compared with five types of data transformation techniques, namely mean and median in monthly data and the rest is in daily data such as binary, cumulative and actual values. Results indicate that data transformation using X-means data splitting in hierarchical clustering outperformed other transformation techniques and more consistent between training and testing datasets based on similarity measures.