Metaheuristic-Based Intelligent Solutions Searching Algorithms of Ant Colony Optimization and Backpropagation in Neural Networks

Abstract

Computation is concerned with the validation of algorithm, estimation of complexity and optimization. This requires large dataset analysis for the purpose of finding the unknown optimal solution. Aside the intricacies in completing tasks, this process is expensive and time inefficient; attaining solutions with conventional mathematical approaches are unrealizable. Search algorithms were advanced to improve solution approaches for optimization problems by finding the possible sets of solution to a particular problem as contained in search space. However, metaheuristic algorithms suggest three solutions to optimization problems on the basis of the application areas in real-life situations including: near optima, the optimal or the best solution. This paper analyses the decision-making processes of two nature-inspired search algorithms namely: Backpropagation search algorithm and ant colony optimization (ACO). The results revealed that, backpropagation search algorithm without ACO training trailed those trained with ACO for MSE, RMSE, RAE and MAPE. Again, forecasts errors estimated in the neural network set-up were smaller due to directional search mechanism of the ACO as against the approach provided in neuro-fuzzy rules set tuning by Rajab and Sharma (Soft Comput 23:921–936, 2017) [1]. There is need to consider metaheuristic algorithms approaches to obtain better solutions or nearest optimal values to the optimization problems in neural networks.