Artificial neural network model based on COOT optimization algorithm

Artificial Neural Networks (ANNs) are an artificial intelligence technique that models the information processing and transmission systems of biological neurons in the brain. In recent years, significant progress has been made in artificial neural network models, and these models have been successfully applied to various real-world problems such as pattern recognition, image processing, and classification. However, a limitation of artificial neural networks is that they can get stuck in local minima during the training phase. This is a consequence of using gradient descent-based techniques and negatively affects the generalization performance of the network. Various metaheuristic algorithms have been used in the literature to improve the performance of the network and address the challenging problem of parameter optimization in artificial neural networks. Metaheuristic algorithms are inspired by nature and have been successfully used in many complex optimization problems. In this thesis, a new hybrid artificial neural network model called COOT-ANN is proposed to achieve parameter optimization of artificial neural networks. The COOT-ANN model utilizes a novel metaheuristic approach called the COOT optimization algorithm to optimize the parameters of the artificial neural network. The COOT optimization algorithm is a swarm intelligence-based metaheuristic algorithm developed by modeling the different movements of coots, waterbirds, in their food foraging on the water. By employing the metaheuristic-based COOT optimization algorithm, the COOT-ANN model avoids getting stuck in local minima during the training phase. In the study, classification processes were performed on wine, breast cancer, iris flower data sets, and sports data set for video classification, and the performance of the proposed model was compared with gradient descent methods. The results of the study demonstrate that the proposed approach achieves high accuracy, cross-entropy, F1 score, and Cohen's Kappa metrics on test datasets compared to Gradient Descent, Scaled Conjugate Gradient, and Levenberg-Marquardt optimization techniques. The thesis work highlights the improved performance of using the COOT optimization algorithm in ANN training for classification problems.