Seismic data evaluation using machine learning algorithms

Abstract

Earthquakes are one of the devastating natural disasters which cause significant damage to property due to their destructive nature. Seismic stations around the globe record data continuously to make it available for research and information purpose. An enormous amount of research has been done in this regard in the past as well but generally, the research is done on the seismic regions only. This identifies that there is limited work done on the data analysis for country-wise seismic data. This thesis specifically analyzes and evaluates collective countrywise seismic data through machine learning algorithms. From a geological perspective, Pakistan is located on three tectonic plates. The historic seismic activity of Pakistan along with its neighboring countries including China and Afghanistan is considered for an efficient evaluation. For an unbiased comparative analysis, two evaluation techniques are considered that include threshold based binary seismic classification and magnitude categorization based on the Mercalli intensity scale for determining magnitude destructive nature. Decision tree, Random forest, XGBoost, Adaboost, and KNN are implemented on three country-wise seismic datasets. Among the five applied algorithms, two algorithms including Random forest and XGB performed exceptionally well in the selected evaluation methods. The proposed evaluation methods can be applied to other natural hazardous data as well to evaluate the performance of applied algorithms on selected evaluation criteria. All the algorithms are compared on the basis of selected comparative metrics that provides an insight into the quality of the algorithm performance on country-wise seismic historic activity.