An Optimized Random Forest Classifier with Incremental Learning Using Fuzzy Approach as a Novel Node Splitting Method

Abstract

Random Forest is an ensemble classifier which operates by decision trees. The prediction of the nodal points and formation of rules for Gini Index and Information Gain as the splitting criteria is analysed.  The demerits of Gini Index is that it is not compatible for more partitions and the drawback with Information Gain is that it is not compatible for more distinct values. These setbacks can be ratified by changing the splitting criteria which will yield a better split and in turn the accuracy of Random Forest can be enhanced. In this paper, we propose a new model with a fuzzy approach in Random Forest Algorithm to surmount the dearth of both Gini Index and Information Gain. Fuzzy approach also facilitates the adaptability of the data set.