Profiling the Spatial and Temporal Properties of Earthquake Occurrences Using ST-DBSCAN Algorithm

Earthquakes are a special phenomenon that cannot be prevented or avoided. However this natural disaster can be mitigated to reduce its risk of impact. Earthquakes often occur sequentially and converge at a location and time of occurrence. Data having these characteristics can be analyzed precisely using density-based clustering. This paper presents knowledge discovery from the spatial and temporal properties of earthquake data set by applying ST-DBSCAN algorithm, a spatial temporal density-based clustering. Data clustering is done by considering the location of earthquake coherently with its time of occurrence. The main goal of data clustering is to identify the earthquake prone areas and recognize the pattern of earthquake activity in Nusa Tenggara Barat (NTB), a province in the southeast of Indonesia. Research was conducted in four main steps. First is determining the algorithm hyperparameters which include spatial threshold based on the earthquake location, temporal threshold based on the date of occurrence, and minimum number of a cluster density. The Silhouette index is used to validate the results. Second is data clustering with ST-DBSCAN algorithm. Third is cluster profiling to analyze data distribution spatially and temporally. Fourth is recognizing the pattern of earthquake occurrences. Research shows a one year of data has the optimal value of algorithm hyperparameter with spatial threshold is 0.25 degree (27.75 km), temporal threshold is 3 days, and minimum points of cluster density is 9 occurrences. Data profiling identified there are three districts of NTB were prone to the earthquakes, i.e. East Lombok, North Lombok, and Central Lombok, with the highest earthquakes occurring between July 28 to September 27, 2018. Additionally, research revealed that throughout that time frame, stationary patterns were mostly observed.