Alternative Data Mining Techniques for Predicting Tropical Cyclone Intensification

Predicting the future behavior of tropical cyclones is a problem of great importance for the atmospheric science community with concrete applications. Researchers understand enough about modeling storm systems to predict their track, but forecasting their future intensity remains elusive. In the present study, we formulate tropical storm intensification prediction as a supervised data mining problem; the objective being to produce accurate early warnings with respect to changes in wind speed of a particular storm. We examine two alternative approaches to discover classification rules on current hurricane data: particle swarm optimization and class association rules. Particle swarm optimization employs a population based search method to optimize a rule quality function and discover patterns in the data. Classification with association rules finds sufficiently supported trends in the data and transforms this knowledge into sets of rules. We examine both approaches in detail, present our findings and discuss their impact.