ANALYSIS OF ACCURACY IMPROVEMENT IN RANDOM FOREST USING PRINCIPAL COMPONENT ANALYSIS (PCA)

Abstract

Decision tree is used to classify a data that still does not know its class to existing classes. The data testing path is the first step that the root node goes through and finally the leaf node will predict the class for the data that has been concluded. Random Forest cannot be relied on for data types that have different categorical variables and therefore needs to be improved in the classification process, this is influenced by differences in the value of the variable. Therefore a method is needed to reduce features that are less relevant to the process of determining accuracy in the classification of the Random Forest method. In research conducted on the PCA + Random Forest classification model, using the Water Quality Status Dataset that has been simplified into 5 attributes, 4 classes and 117 instances with an accuracy rate of 91.43% with a classification error rate of 8.57%. Based on the test results from the four classification models, it can be concluded that the success of the PCA can be used as a reference to improve the accuracy performance of the Random Forest classification model