Earthquake Alert Prediction Using Machine Learning Models for Resilient Infrastructure

Earthquake early warning systems are necessary to reduce deaths and damage to buildings by sending out alerts before earthquakes happen. This study investigates the application of machine learning algorithms, specifically Neural Networks, AdaBoost, Naive Bayes, k-Nearest Neighbours (kNN), and Gradient Boosting, for predicting earthquake alerts based on parameters such as magnitude, depth, Community Disaster Index, Modified Mercalli Intensity (MMI), and seismic significance (sig). Normalisation and class balancing were used to preprocess the dataset and make the model more accurate. The Orange data mining environment used the following measures to quantify performance: Classification Accuracy, Precision, Recall, F1-score, ROC-AUC, and Matthews Correlation Coefficient. The tests demonstrated that ensemble models worked far better than standard classifiers. AdaBoost was the best at making predictions, with 100 % accuracy and recall. Gradient Boosting came in second, with CA and ROCAUC values of 0.97. This means that it is good at finding earthquake circumstances that are very dangerous. The findings indicate that ensemble-based machine learning techniques may enhance the responsiveness and reliability of seismic early warning systems. This would assist places that are likely to be affected by disasters make decisions faster and be better prepared for a resilient infrastructure.