Neural Network Architecture Selection Using Particle Swarm Optimization Technique

Finding the best structure of ANN to minimize errors, processing and search time is one main objectives in AI field. In order achieve prediction with a high degree accuracy short time, an enhanced PSO-based selection technique determine optimal configuration for artificial neural network has been proposed this paper. To design maximize prediction, it necessary identify hyperparameter values precision. PSO 2-D space employed select hyperparameters construct where used as decision-making model learning model. The suggested was number hidden layer units per layer. evaluated using chemical dataset. result testing displayed MSE equal 3.9% relative error between expected output actual target less than 1.6%. results comparison showed thatthe approach could predict infinitesimal error, outperforming existing terms ratio.