Earthquake Damage Prediction Using Random Forest and Gradient Boosting Classifier

Sourav Pandurang Adi,Keshav Vaidyanathan Bharadwaj,Vivek Bettadapura Adishesha,Abhinav Narayan

doi:10.11648/j.ajbes.20200603.14

Sourav Pandurang Adi, Keshav Vaidyanathan Bharadwaj + Show 2 more

Open Access

https://doi.org/10.11648/j.ajbes.20200603.14

Copy DOI

Abstract

Earthquake is a major natural disaster that causes casualties in millions and leaving many more in trauma. Analyzing the consequences of such consequences gives one a better stand-in for potential catastrophe occurrences. It is important to establish a methodology that can assist in forecasting these earthquakes, as they can help prevent the severity of the damage. This paper discusses a machine learning model that can predict the damage grade severity caused by life-threatening earthquake that hit Nepal in the year 2015. The dataset is derived from the live competition hosted by Driven Data. The data was collected through the surveys conducted by the Kathmandu Living Labs and the Central Bureau of Statistics, which operates under the National Planning Commission Secretariat of Nepal. To accomplish the defined goal, we used the Random Forest Classifier and Gradient Boosting Classifier. The Random Forest Classifier algorithm demonstrated in this study was outperformed by the Gradient Boosting Classifier. With necessary parameter tuning using the Random Forest Classifier, the F1-Score achieved was 72.95%. The next technique was to perform winsorization on some attributes to handle outliers which improved the F1-score to 74.33% along with gradient boosting classifier. The last techniqueinvolved only hyper-parameter tuning with gradient boosting classifier achieved the best F1-Score of 74.42%.

Highlights

Earthquakes almost always occur on faults and on the surfaces of the earth where one side is rising in relation to the other
Earthquakes occur on faults, previously identified by geological mapping, which shows that motion across the fault has occurred in the past
Reports at the time of the quake described the number of trekkers and climbers at American Journal of Biological and Environmental Statistics 2020; 6(3): 58-63 base camp as up to 1000

Summary

Introduction

Earthquakes almost always occur on faults and on the surfaces of the earth where one side is rising in relation to the other. We have used dataset given by driven data [19], performed EDA using Tableau [12], and developed a machine learning model that is capable of predicting the damage grade severity to the buildings caused by the earthquake [1, 2, 4, 9, 16, 17]. Rapid assessment of damage severity to the buildings is an essential post-event recovery To achieve this Sujith Mangalathu (2019) [4] et al evaluated the possibility of using various machine learning techniques such as K-Nearest Neighbors, Random Forests, Decision Trees, etc. With the available seismological data, building damage was linked by reviewing the seismotectonic setting of Nepal, earthquake rupture process, aftershock data which was provided by the U.S Geological Survey (USGS). The authors use Random Forest algorithm to produce more reliable maps

Dataset

Data Preparation

Exploratory Data Analysis

Training and Testing the Model

Results and Evaluation

Conclusion