Using Random Forests and Fuzzy Logic for Automated Storm Type Identification

This paper discusses how random forests, ensembles of weakly-correlated decision trees, can be used in concert with fuzzy logic concepts to both classify storm types based on a number of radar-derived storm characteristics and provide a measure of “confidence” in the resulting classifications. The random forest technique provides measures of variable importance and interactions, as well as methods for addressing missing data, suggesting fruitful ways to transform the input data and to structure the final classification algorithm. N-fold cross-validation is used as the basis for tuning the algorithm parameters.