Data mining methods for analyzing and modeling spatiotemporal data

The advance in modern data acquisition and management techniques result in a big and dynamic collection of spatiotemporal data. These emerging spatiotemporal data brings in both challenges and opportunities for understanding the complex temporal/geographic processes and phenomena from different fields. Methods targeted at learning from spatiotemporal need to handle the huge amount of features and complex regularities created by the variations among spatial,temporal and multivariate spaces. In this dissertation, we developed and applied learning models for real word spatiotemporal data sets according to different analytic purposes, such as: 1) Geospatial Discriminative Pattern and Hotspot Optimization Tool for visualizing and clustering crime hotspots, 2) Streaming Feature Selection and Most-Correlated Set for identifying precursors of extreme precipitation clusters, 3) Multiple Markov Boundaries for forecasting tornado, and 4) Hierarchical Pattern Learning for predicting extreme rainfall events.