K-Means Using Dynamic Time Warping For Clustering Cities in Java Island According to Meteorological Conditions

Java is an island in the tropical region and it is the most densely populated island in Indonesia. Analyzing the meteorological data in Java is beneficial to find information that might be a sign of climate change which influences the population and environment. This paper uses daily time series meteorology data from 2010 to 2022 containing variables minimum temperature, maximum temperature, average temperature, humidity, sunshine duration, rainfall, average wind speed, and maximum wind speed. It implements the K-Means algorithm using dynamic time warping to cluster cities in Java Island and nearby based on the meteorological conditions. The experiments are run to cluster the data with the number of clusters $k=\{2,3,4,5\}$ K-Means using dynamic time warping produces an average silhouette score is around 0.26 obtained when the number of clusters $k=2$. K-Means using dynamic time warping outperforms Fuzzy C-Means and Intelligent K-Means. The clustering result for $k=2$ shows that Pasuruan, Nganjuk, and Malang belong to Cluster 1. Those cities have a trend of lower average temperature, higher precipitation, and higher humidity than cities in Cluster 2. The general trend in Cluster 1 shows there exists a slightly increasing temperature and decreasing wind speeds. The capitals of provinces in Java, i.e., Surabaya, Semarang, Jakarta, Bandung, Yogyakarta, and Serang, are in the same clusters together with Cilacap, Tegal, Banyuwangi, Bogor, Sumenep, Gresik, Majalengka, Tangerang, and Tangerang Selatan. An annual increasing average temperature happens in Serang, Nganjuk, and Gresik. The common trend of Cluster 1 and Cluster 2 is consistently decreasing wind speeds which are possible to be early warnings as signs of climate change.