K-Means and K-Medoids Algorithm Comparison for Clustering Forest Fire Location in Indonesia

Abstract

Forest fires are the most common cause of deforestation in Indonesia. This condition has a negative impact on the survival of living things. Of course, this has received special attention from various parties. One effort that can be made for prevention is to group these points into areas with the potential for fire using the clustering method. In this research, a comparative study of the clustering algorithm between K-Means and K-Medoids was conducted on hotspot location data obtained from Global Forest Watch (GFW). Besides that, important variables that affect the clustering process are also analyzed in terms of feature importance. There are nine important variables used in the clustering process, of which the Acq_time variable is the most important. The cluster quality of both algorithms is evaluated using the silhouette coefficient (SC). Both algorithms are capable of producing strong clusters. The best number of clusters is six clusters. The K-medoids algorithm is better at grouping data than K-means.

Analyzing the Effect of Similarity Functions on Partitioning Around Medoids Algorithm for Mapping DHF Disease in North Sumatra

“…Subjective criteria for judging strong, weak, or not-found clusters by 𝑆𝐼 are previous research criteria [22]. The criteria are shown in Table 3.…”