Amalgamation of Non-dominated Sorting Genetic Algorithm-II with Random Forest for Improved Prediction of Protein-Protein Interaction

In this paper, we present a novel hybrid classifier aimed at improving the accuracy of protein–protein interaction (PPI) predictions. This classifier combines the multi-objective Non-dominated Sorting Genetic Algorithm-II (NSGA-II) with Random Forest. The primary objective is to generate a variety of accurate decision trees while simultaneously determining the optimal number of trees needed. In our study, we first utilize a natural vector method to represent proteins as 60-dimensional vectors. Instead of employing traditional approaches, we then implement an evolutionary multi-objective strategy. This allows us to generate various training sets with varying numbers of samples and features for constructing each decision tree. Additionally, the solutions acquired from the Pareto-optimal fronts help identify the necessary number of training sets required to construct the Random Forest. As a result, the performance of the Random Forest is enhanced, leading to improved prediction accuracy. Our findings indicate that the suggested framework, featuring an optimal number of classifiers, significantly prevails over other algorithms employing different classifiers.