Machine Learning for Characterizing Climate-related Disasters (Student Presentation, Group 20)

This is a STAT 451 class project presentation

by Eliot Kim, Jonathan Lala, and Noah Simandl

This presentation is shared with the students' permission.

Abstract:

Predictions of disaster impact are critical in mitigating human and material losses. Accurate predictions are especially necessary in the coming years due to global warming, which has increased the severity of natural disasters. Ma- chine learning algorithms provide a powerful tool to model the complex relationships among the natural, social, and economic variables which dictate the impact of disasters. The aim of this project is to compare the performance of prevalent machine learning algorithms in disaster impact prediction. Models used are decision trees, k-nearest neighbors, and artificial neural networks. Training data was provided by EM-DAT, an international disasters database. Feature selection and transformation was conducted to create suitable input data for the models. Based on 59 in- put features, each model outputs the expected amount of damage on a categorical scale from 0 to 11. All models showed potential to provide accurate disaster impact pre- dictions, with the k-NN and ANN resulting in higher performance than the decision tree. Larger datasets and additional model tuning would likely result in improved performance.