Red-billed blue magpie optimization for training feedforward neural networks

The input layer, hidden layer, and output layer constitute the feedforward neural networks (FNNs) with unidirectional propagation and no feedback loop connection. The FNNs exhibit pivotal properties of nonlinear fitting ability, flexible structure, multitasking adaptability, easy implementation, strong compatibility and composability. The FNNs can independently tackle classification and regression tasks and support the construction of complex networks. The red-billed blue magpie optimization (RBMO) is constructed on the collective collaborative exploration and sophisticated hunting proficiency of the red-billed blue magpies. The RBMO emulates searching for food sources, chasing and attacking prey, and hoarding superfluous food to balance extensive global exploration and meticulous local exploitation, and to ascertain the fittest solution. In this paper, the RBMO is proposed to train FNNs, the motivation is to measure the dissimilarity between anticipated and actual outputs, quantify the training efficiency and classification precision of the sample datasets, and fabricate the connection weights and biases. The RBMO exhibits the distinctive advantages of extensible adaptability and flexibility, accelerated convergence speed, heightened calculation accuracy, decreased control parameters, remarkable robustness and stability, and diminished computational complexity. Seventeen sample datasets are utilized to authenticate the applicability and practicality, the RBMO is compared with SFOA, FLO, APO, HEOA, EGO, BKA, WO, HO, IAO, NRBO, ETO and PO. The experimental results demonstrate that the RBMO can dynamically alternate between exploration and exploitation to strengthen stability and robustness, intensify training efficiency and classification precision, promote convergence speed and solution quality.