An improved artificial neural network based on human-behaviour particle swarm optimization and cellular automata

•A local version of HPSO with CA is proposed to improve convergence performance.•HPSO-CA is combined with ANN to prevent ANN from trapping in local minima.•ANN-HPSO-CA is proved to be effective to train ANN’s connectivity weights.•The proposed ANN-HPSO-CA shows a competitive performance on the tested datasets. Back-Propagation (BP) neural network, as a powerful and adaptive tool, has led to a tremendous surge in various expert systems. However, BP model has some deficiencies such as getting trapped in local minima and premature convergence. These weaknesses can be partly compensated by combining the ANN with Evolutionary Algorithms (EAs), i.e., at the same time, EAs also sufferred from their own characteristics, such as premature convergence in Particle Swarm Optimization (PSO). To gain a better trained weights in EAs-ANN, this paper proposes an improved ANN model based on HPSO and Cellular Automata (CA), which is called ANN-HPSO-CA. Firstly, to balance global exploration and local exploitation better and prevent particles from trapping in local optima, CA strategy is involved in HPSO algorithm, which is denoted as HPSO-CA. Then, the proposed HPSO-CA algorithm is combined with ANN to prevent ANN from trapping in local minima. Finally, to validate the performance of ANN-HPSO-CA, 15 benchmark complex and real-world datasets are used to compare with some well-known EA-based ANN models. Experimental results confirm that the proposed ANN-HPSO-CA algorithm outperforms the other predictive EA-based ANN models. The numerical comparison results will provide useful information and references for any future study for choosing proper EAs as ANN training algorithms. In addition, ANN-HPSO-CA algorithm provides a good theoretical basis for an expert system with good convergence and robustness.