A double weighted fuzzy gamma naive bayes classifier

摘要

Classifiers based on Gamma statistical distribution can be found in the scientific literature, but they assume the collected data doesn't present any errors. However, in some cases, information precision can not be guaranteed, then the fuzzy approach is convenient. Several methods found in the literature are not able to ponder the specific contribution of each class and/or feature for the classification tasks. This paper presents a proposal of a new classifier named Doubled Weighted Fuzzy Gamma Naive Bayes network (DW-FGamNB). This new classifier uses two types of weights in order to allow users to ponder the real contribution of each class and feature in the classification task. The theoretical development is presented, as well as results of its application on simulated multidimensional data using Gamma statistical distribution. A comparison among DW-FGamNB, Fuzzy Gamma Naive Bayes classifier, classical Gamma Naive Bayes classifier, Naive Bayes classifier, DecionTree-Naive Bayes, Decision Tree C4.5, Logistic Regression, Multilayer Perceptron Neural Network, Adaboost-M1, Radial Basis Function Network and Random Forest was performed. The results obtained showed that the DW-FGamNB produced the best performance, according to the Overall Accuracy Index, Kappa and Tau Coefficients, and diagnostic tests.