Because of low resolution of GFS data, detecting a possible occurrence of hurricane can be performed in two steps: picking a possible grid from GFS and examining further with high resolution data.

1. Setting up a problem
  * To classify the grid that has hurricane.
  * To suggest a probability that hurricane exist on each grid. 
  * To predict a possible (exact) location of hurricanes 

2. Preprocessing of Data
  * Unbalanced data handling
  * Target setup corresponding problem 

3. Large Data Processing 
  * Vector Quantization 
  * Chunking (SVM)

4. Testing and selecting classification method

Optimized Cutting Plane Algorithm for Large-Scale Risk Minimization

Incremental and Decremental Support Vector Machine Learning