Multi-Layer Perceptron Training Optimization Using Nature Inspired Computing

Although the multi-layer perceptron (MLP) neural networks provide a lot of flexibility and have proven useful reliable in wide range classification regression problems, they still limitations. One most common is associated with optimization algorithm used to train them. The commonly training method stochastic gradient descent backpropagation (or for short) because it mathematically tractable (given that activation functions are differentiable). However, not guaranteed find globally optimal set weights biases. As result, MLP often incapable obtaining desirable solution problem. Clonal selection algorithms (CSA) procedures effectively explore complex large space values near global optimum. Consequently, CSA can be solve problem networks. This paper presents novel implementation architectures real-world problems such as breast cancer diagnosis, active sonar target classification, wheat flower classification. biases will significantly increase accuracy MLP. performance our proposed approach compared other popular methods: genetic (GA), ant colony (ACO), particle swarm (PSO), Harris hawks (HHO), moth-flame (MFO), pollination (FPA), (BP). comparison benchmarked using five datasets: Iris Flower, Sonar, Wheat Seeds, Breast Cancer Wisconsin, Haberman’s Survival. Comparative study results illustrate improvements gained by over methods, hence considered competitive when solving applications various disciplines.