Training Multi-Layer Perceptron Using Population-Based Yin-Yang-Pair Optimization

Abstract

This work proposes the Population-Based Yin-Yang-pair Optimization (PYYPO) algorithm and its implementation for successfully training an multi-layer perceptron (MLP) neural network. The proposed PYYPO trainer is then used on five healthcare-related classification datasets: breast cancer, diabetes, liver disorders, Parkinson disease, and vertebral column. The results of the proposed trainer are compared with four other algorithms: Particle swarm optimization, genetic algorithm, cuckoo search, and moth flame optimization algorithms. Lastly, it is shown in the findings that the proposed trainer achieves better scores on most datasets when compared to other algorithms on one performance metrics, namely classification accuracy.