Improving the Performance of Naïve Bayes Algorithm by Reducing the Attributes of Dataset Using Gain Ratio and Adaboost
Fanny Ramadhani, Al-Khowarizmi Al-Khowarizmi, Indah Purnama Sari
Abstract
Naive Bayes is one of the most well-known data mining algorithms for classification. Naive Bayes is a simple and effective learning theory that does not need various parameters. However, Naive Bayes also has its drawbacks. The obstacle faced by naive bayes is its performance and accuracy decreases when the data to be classified contains a large number of features and dimensions. In this study, the author will improve the accuracy and performance of naive bayes by using the gain ratio in reducing attributes and boosting using Adaboost. The gain ratio is used to select and reduce features in the dataset. And adaboost approach for ensemble learning to improve the performance of Naive Bayes in classifying the data set of clinical heart failure records. Naive Bayes will be used as a base learner. Validation is carried out using the 10-fold Cross Validation approach. While the confusion matrix is used to test the accuracy. The naive bayes performance before combined with the gain ratio and adaboost is the TPR value of 0.29, the TNR value of 0.96, the NPV value of 0.94, the FNR value of 0.71, the FOR value of 0.06 and the naive bayes accuracy. 0.91. After combining naive bayes with gain ratio and Adaboost, naive bayes performance improves. Namely, the accuracy of Naive Bayes performance increased to 0.94 and the TPR, TNR and NPV values also increased to 0.571, 0.975 and 0.9642.