Fuzzy decision tree using soft discretization and a genetic algorithm based feature selection method

In data mining, decision tree learning is an approach that uses a decision tree as a predictive model mapping observations to conclusions. The fuzzy extension of decision tree learning adopts the definition of soft discretization. Many studies have shown that decision tree learning can benefit from the soft discretization method leading to improved predictive accuracy. This paper implements a Fuzzy Decision Tree (FDT) classifier that is based on soft discretization by identifying the best “cut-point”. The selection of important features of a data set is a very important preprocessing task in order to obtain higher accuracy of the classifier as well as to speed up the learning task. Therefore, we are applying a feature selection method that is based on the ideas of mutual information and genetic algorithms. The performance evaluation conducted has shown that our FDT classifier obtains in some cases higher values than other decision tree and fuzzy decision tree approaches based on measures such as true positive rate, false positive rate, precision and area under the curve.