Improved Prediction of Eukaryotic Protein Subcellular Localization Using Particle Swarm Optimization of Multiple Classifiers

Abstract

This paper proposes an improved prediction of protein subcellular localization in eukaryotes based on multiple classifiers' prediction results. A modified Particle Swarm Optimization was used to screen the prediction results and determine the optimal answer for protein locations. By weighting the probabilities from each classifier's result, the final set of prediction results were formed. The location with highest probability in each final set was selected and checked with the given class labels. The number of correct answers was calculated to evaluate the PSO performance. Then, the results were compared with other classifiers in terms of accuracies. The overall accuracy achieved by the proposed PSO method is 75.63%.