Backpropagation neural network based on local search strategy and enhanced multi-objective evolutionary algorithm for breast cancer diagnosis

• Main Authors: Ashraf Osman, Ibrahim, Siti Mariyam, Shamsuddin, Abdulrazak, Yahya Saleh, Ahmed, Ali, Mohd Arfian, Ismail, Shahreen, Kasim
• Format: Article
• Language: English
• Published: Insight Society 2019
• Subjects: QA Mathematics; QA75 Electronic computers. Computer science; QA76 Computer software; TK Electrical engineering. Electronics Nuclear engineering
• Online Access: http://umpir.ump.edu.my/id/eprint/25082/1/Backpropagation%20neural%20network%20based%20on%20local%20search%20strategy.pdf; http://umpir.ump.edu.my/id/eprint/25082/; https://doi.org/10.18517/ijaseit.9.2.4986; https://doi.org/10.18517/ijaseit.9.2.4986
• Institution: Universiti Malaysia Pahang

The role of intelligence techniques is becoming more significant in detecting and diagnosis of medical data. However, the performance of such methods is based on the algorithms or technique. In this paper, we develop an intelligent technique using multiobjective evolutionary method hybrid with a local search approach to enhance the backpropagation neural network. First, we enhance the famous multiobjective evolutionary algorithms, which is a non-dominated sorting genetic algorithm (NSGA-II). Then, we hybrid the enhanced algorithm with the local search strategy to ensures the acceleration of the convergence speed to the non-dominated front. In addition, such hybridization get the solutions achieved are well spread over it. As a result of using a local search method the quality of the Pareto optimal solutions are increased and all individuals in the population are enhanced. The key notion of the proposed algorithm was to show a new technique to settle automaticly artificial neural network design problem. The empirical results generated by the proposed intelligent technique evaluated by applying to the breast cancer dataset and emphasize the capability of the proposed algorithm to improve the results. The network size and accuracy results of the proposed method are better than the previous methods. Therefore, the method is then capable of finding a proper number of hidden neurons and error rates of the BP algorithm.