An optimized phishing detection model using hybrid feature selection and a fine-tuned narrow neural network with dynamic jaya optimization to overcome cyberthreats

Abstract The frequency of phishing assaults threatens the occurrence of online transactions. Criminals engaged in phishing utilize misleading strategies, such as fake URLs, to obtain sensitive information from consumers through media such as e-mails and texts. More efficient phishing detection techniques that use classifiers combining hybrid approaches of feature selection techniques and deep learning must be developed to handle attacks. The proposed work provides a unique method for improving phishing detection that combines a hybrid feature selection strategy with a fine-tuned narrow neural network (FTNN) and Dynamic Jaya optimization algorithm. The model works in two phases: a hybrid feature selection algorithm works for filtering the salient features and a fine-tuned narrow neural network architecture with and without optimization is suggested for classifying phishing and legitimate urls. The process starts by choosing a subset of 20 features from a total of 48 attributes. The model specifically uses chi-square and Kruskal–Wallis statistical feature ranking algorithms. The FTNN with Dynamic Jaya optimization performs best when large datasets are used for feature selection. The results demonstrate the superior performance of the proposed hybrid feature selection technique along with the Dynamic Jaya optimized FTNN architecture. It outperforms conventional methods that rely on all the features, resulting in higher accuracy than existing models. A higher accuracy of 97.5% has been achieved with the integration of hybrid feature selection with FTNN, compared to other traditional methods. This research provides a practical and desirable solution for enhancing hybrid feature selection along with a fine-tuned narrow neural network (optimized with Dynamic Jaya) in phishing detection. Hence, this work proposes an innovative hybrid feature selection approach, coupled with a fine-tuned neural network with enhanced recognition performance by the Dynamic Jaya algorithm, to present an efficient, accurate, and scalable identification of phishing compared to traditional methods. These innovations underscore the novelty of feature selection, light neural network, and optimization that in total enhances performance and applicability in tackling phishing.