Enhancing Random Forest Performance through Optimal Instance Subset Selection using Genetic Algorithm

Abstract

The exponential growth of data in the information system has led the concept of big data. High dimensional data imposes serious challenges to the useful information. It limits the traditional machine learning algorithms and techniques to preprocess, learn, and analyze the data. The optimal subset selection is an effective solution. Thus, feature and instance selection are used for data reduction. In this paper, we have chosen instance subset selection to increase the accuracy of data analysis. The study is to propose an approach to select the optimal set of instances using Genetic Algorithm and is also aimed to optimize the Random Forest using GA for the boosted accuracy in the classification. The study introduced an approach that uses Genetic Algorithm to select the optimal instance subset. GA is also implemented with the Random Forest algorithm to select the best parameters that are stored in the DEAP library for the accurate classification. The experimentation results show that Genetic Algorithm effectively selects the best instance subset, and the combination of Random Forest and Genetic Algorithm outperforms with 85.8%. This method not only shows how evolutionary techniques can be used for data-driven optimization, but it also makes the model more efficient by decreasing data redundancy.  The findings provide credence to the idea that combining Genetic Algorithms with more conventional machine learning techniques can improve performance.