Predicting Impact of Natural Calamities in Era of Big Data and Data Science

Session G2: Data Mining for Environmental Sciences (s-DMTES IV)

Every year thousands of lives and millions of dollars are lost during natural catastrophic events. Data scientists have recognized the value of employing data mining techniques for evaluating these events to assist in understanding trends, predicting future disasters, and assessing vulnerability of populations. The goal of such analysis is to prepare governments for emergency response and relief efforts as well as to formulate strategies for future disaster mitigation. In addition, increasing pressures from a growing world population further emphasize the need for governments to seek viable solutions for balancing human needs with environmental constraints. Identifying populations vulnerable to environmental calamities includes analysis of past events to identify patterns which may identify at-risk populations and improve predictions of future events. Past analytical techniques used common statistical methods; however, recent developments in the field of data science and Big Data technologies have enabled data scientists to apply predictive and descriptive techniques accurately and economically. In this work, both supervised and unsupervised learning techniques will be applied to natural disaster data extracted from the Emergency Events Database (EM-DAT) for North America, Central America, and the Caribbean. Unsupervised, descriptive analytics were performed using the density based clustering algorithm DBSCAN to identify notable patterns in the disaster data. A supervised, predictive model was built using an artificial neural network to predict the potential monetary impact of natural calamities based on region, country, and natural disaster type.