Calculation of Neural Network Weights and Biases Using Particle Swarm Optimization

Various machine learning techniques and algorithms have been used to address, and are still being used to tackle, several real-world issues. One technique that has been extensively employed to address a variety of issues is the usage of neural networks. Neural networks can be used to classify data and to calculate regression coefficients. Backpropagation is the cornerstone of neural network training. The process of iteration involves changing the weight of a neural network in response to the rate of error observed in the preceding epoch. The error rates can be reduced and the applicability of the model increased, both of which will increase the model’s dependability. Artificial neural networks are commonly trained using the backpropagation approach, also known as backward propagation of mistakes. This technique aids in figuring out a loss function’s gradient for every network weight. The backpropagation method divides the dataset into training and testing sets. The neural network is assisted in performing exploration and exploitation using a variety of techniques. Among them are algorithms with biological inspiration. By using a different approach, bio-inspired computing can be distinguished from other traditional algorithms. Simple rules and individual life forms or swarms of individuals that adhere to those rules make up the ideology of bio-inspired computing. These living things, also referred to as agents, develop over time and advance with fundamental imperatives. This approach can be categorized as bottom-up or decentralized. In this paper, a neural network is created using weights and biases determined using the swarm’s individual particles. To compare a few parameters between the particle swarm optimization and backpropagation in neural networks, the Pima Indian diabetes dataset is employed.