Learning crisis management information system from open crisis data using hybrid soft computing

Abstract

Crisis management is being dealt extensively with reliable data sources that are being collected continuously by na- tional and local authorities. These open data creating a radical opportunity for the successful prediction of crisis. At the same time, it poses some challenges to the researchers for its heterogeneity, real time and massive data. Recently, soft computing based data mining plays a very vital role as a source of innovation in crisis information management, in obtaining more valuable information. In this paper, we propose to use neural network based forecasting for obtaining post crisis scenario. Further, we use hybrid soft computing methodologies by combining attribute selection with PSO (Particle Swarm Optimization), GA (Ge- netic Algorithm) and EA (Evolutionary Algorithm); hybrid classification by fuzzy-rough VQNN (Vaguely Quantified Nearest Neighbor) and MLP (Multilayer perceptron neural network). We use three open datasets such as: rival crisis, dyadic and global terrorism data collected from World Bank for our experimentation. Finally, we conclude with encouraging results obtained from EA based VQMLP that provides best prediction in terms of various performance measures considered, for all crisis scenarios which will help us in understanding to take future course of action in such a situation.