Modeling Earthquake Damage Grade Level Prediction Using Machine Learning and Deep Learning Techniques

Abstract

Predicting earthquake damage grade level has been a much needed and important research area, where the later instances of the destructive damage can be speculated. This paper deals with the research work on Modeling Earthquake Damage grade level, using the dataset available in the DrivenData platform, which consists of 39 features having values assigned to each one of the rows. Training the models to predict the damage caused by an earthquake in Nepal is done using machine learning and deep learning approaches such as Random Forest Classifier (RFC), Logistic Regression, K-Nearest Neighbor, Decision Tree and Artificial Neural Networks and their performance is evaluated based upon their evaluation metrics. Highest F1 score and minimum loss are obtained in Random Forest Classifier method. RFC helps in general level classification for predicting earthquake damage grade level effectively with a F1-score of 84.46%. However, it is observed that 29 important features obtained during RFC training are enough to yield a F1-score of 74.36%.