A Collection of Contemporary Research Articles in Electronics, Communication & Computation- Book ISBN- 978-81-948050-7-6

Authors: T. Mathi Murugan, Dr. E. Baburaj

Abstract: The utilization of multilayer perception and feed forward network has limitations in the neural network, such as multi-layering and linear threshold unit for different engineering applications. Thus, higher-order neural networks were beneficial in performing non-linear mapping, which utilises the input units with the single layer for conquering the limitations of the conventional neural networks. The paper utilises a higher-order neural network known as the improved Pi-Sigma neural network coupled with the Barnacles Mating Optimization (BMO) algorithm and Black Widow Optimization (BWO) algorithm for developing an effective hybrid training algorithm for classification process with global and local searching abilities. For validating the performance of the proposed BMO-IPSNN and BWO-IPSNN algorithm, the algorithm was tested with different types of datasets obtained from UCI machine learning source, and then the algorithm was evaluated with the other existing algorithms such as PSO-PSNN, IPSNN, FFA-PSNN and PSNN. The result from the experiment concludes that the proposed algorithm gains superior performance for classification problems.

Keywords: Bio-inspired algorithms, Optimization, Classification, Neural networks