Hybrid Feature Selection Technique using Filter (FCBF) & Wrapper Method (SFFS) for Discrete Class Data Mining
Feature Selection can extensively increase the scope of classifiers to work on high dimensional data by reducing the variables or features used for classification. A dataset may represent a number of attributes which have little relevance to classification problem associated with it. By choosing a subset of the most useful attributes, classification is best aided to produce most accurate classification results. This paper uses a hybrid feature selection technique using supervised feature selection techniques. The filter method is used to sort the features in order of their ‘usefulness’ and the wrapper method is used for picking out the best subset from these features to simplify classification. The last step is to create a comparison in performance between attributes acquired as a result of the hybrid model and the total set by passing both via the classifier to calculate error and accuracy.
Keywords - data mining, feature selection, fast correlation based filter, sequential forward floating selection, hybrid model, classification, naïve bayes