Richter's Predictor: Modelling Earthquake Damage Using Multi-class Classification Models

Abstract

Natural calamities like earthquakes cause damage to life and property. Estimation of damage grade to buildings is essential for post-calamity response and recovery, elimination of the tedious process of manual validation and authentication of property damage before granting relief funds to people. By considering basic aspects like building location, age of the building, construction details and it's secondary uses, taken from the Gorkha earthquake dataset, this paper explores various multi-class classification machine learning models and techniques for predicting the damage grade of structures. The proposed architecture of the model involves three major steps, Feature Selection, XGBoost Classifier, and Parameter Tuning. The paper presents the results of the experiments with feature engineering, training variations and ensemble learning. The paper delves into the analysis of each model, to understand the reason behind their performance. This paper also infers the agents that play a major role in deciding the seismic vulnerabilities of the buildings. The proposed classifier in the paper provides significant input to understanding earthquake damage and also provides a paradigm to model other natural disaster damage.