Harmful algal blooms (HABs) pose an enormous threat to the U.S. marine habitation and economy in the coastal waters. Federal and state coastal administrators have been devising a state-of-the-art monitoring and forecasting system for these HAB events. The efficacy of a monitoring and forecasting system relies on the performance of HAB detection. We propose a machine learning based spatio-temporal data mining approach for the detection of HAB events in the region of the Gulf of Mexico. In this study, a spatio-temporal cubical neighborhood around the training sample is introduced to retrieve relevant spectral information of both HAB and non-HAB classes. The feature relevance is studied through mutual information criterion to understand the important features in classifying HABs from non-HABs. Kernel based support vector machine is used as a classifier in the detection of HABs. This approach gives a significant performance improvement by reducing the false alarm rate. Further, with the achieved classification accuracy, the seasonal variations and sequential occurrence of algal blooms are predicted from spatio-temporal datasets. New variability visualization is introduced to illustrate the dynamic behavior of HABs across space and time.