Predictive Analysis of Seismic Activity Using Historical and Real-Time Earthquake Data

Abstract

Earthquakes are among the most unpredictable and destructive natural disasters, causing massive loss of life and property across the globe. Accurate seismic prediction has long been a major scientific challenge due to the complex and nonlinear nature of tectonic processes. In this study, machine learning techniques are applied to analyse historical earthquake data from California to predict the magnitude and probability of future seismic events. The dataset used consists of earthquake records with a magnitude of 3.0 or higher, including parameters such as latitude, longitude, depth, number of seismic stations, and time of occurrence. Various machine learning algorithms — including Linear Regression, Multiple Linear Regression, Decision Tree Regressor, K-Nearest Neighbours (KNN), Support Vector Machine (SVM), Naive Bayes, and K-Means Clustering — were implemented and compared to evaluate their predictive performance. The results demonstrate that machine learning models can effectively capture hidden patterns within seismic data and provide reliable magnitude predictions. Among the tested models, regression-based approaches and SVM showed the best accuracy and consistency. This research highlights the potential of data-driven models in enhancing earthquake forecasting systems, supporting early warning mechanisms, and contributing to disaster risk reduction.