Ameliorating Performance of Random Forest using Data Clustering

Random Forest is one of the most popular supervised learning ensemble methods in machine learning. Random Forest engenders a set of random trees and considers majority voting techniques to classify known and unknown data instances. In Random Forest, decision tree induction is used as a baseline classifier. Decision tree is a top-down divide and conquer recursive algorithm that applies feature selection technique to select the root/best feature e.g. ID3 (Iterative Dichotomiser 3), C4.5 (an extension ID3), and CART (Classification and Regression Tree). In this paper, we have proposed a new approach to improve the performance of Random Forest classifiers using clustering techniques. This proposed idea can be applied for Big Data mining. First, we have clustered the data into several clusters using K-Means Clustering and then applied the Random Forest technique in each cluster. We have tested the proposed idea with existing classical Random Forest technique and found the proposed Random Forest technique performs better than traditional Random Forest algorithm on five datasets taken from UCI Machine Learning Repository.