Intelligent classification with marine predators algorithm and probabilistic neural networks.

The probabilistic neural network technique is a popular data mining process that is used for addressing a variety of classification, prediction, and pattern recognition challenges. One strategy to increase the accuracy of classification is to modify the probabilistic neural network classifier’s weights using optimization techniques. Metaheuristic algorithms have demonstrated their robustness in addressing a variety of engineering challenges. As a result, multiple researchers have utilized metaheuristic algorithms to improve the search process in order to train artificial neural networks in recent years. In this work, the marine predator metaheuristic algorithm is applied. Eleven benchmark classification datasets are used to assess how well the marine predators algorithm performs in adjusting the probabilistic neural network parameters (weights and biases). The proposed method (MPA-PNN) was compared with probabilistic neural network along with three additional strategies from the literature: african buffalo optimizer, hill climbing, and coronavirus herd immunity algorithms. The findings demonstrate that integrating MPA significantly enhances the classification accuracy of the standard PNN. When benchmarked against three prominent metaheuristic-based PNN hybrids from recent literature–namely CHIO-PNN, ABO-PNN, and B-HC-PNN–the proposed MPA-PNN model achieved superior or competitive accuracy on the majority of the 11 UCI datasets evaluated, attaining the highest average accuracy of 91.047%. Furthermore, the results indicate that MPA-PNN exhibits faster and more stable convergence compared to the baseline PNN. While these findings are promising within the specific niche of metaheuristic optimization for PNNs, we acknowledge that further validation against a broader set of contemporary classifiers, such as gradient boosting machines, is a necessary direction for future work to fully establish its generalizability.