Detection of phishing attacks using probabilistic neural network with a novel training algorithm for reduced Gaussian kernels and optimal smoothing parameter adaptation for mobile web services

The rapid escalation of smartphones and online transactions increases the rate of phishing attacks that exploit the user credentials for fraudulent financial gains. The existing detection methods suffer from low detection accuracy and high false positive rate (FPR). In this study, the probabilistic neural network (PNN) with a novel training algorithm is used for detecting phishing attacks. A novel fuzzy dense K-modes (FDKM) clustering algorithm is proposed for obtaining the Gaussian kernels in pattern layer. Moreover, the proposed optimisation procedure called modified harmony search with generation regrouping (MHS_GR) finds the optimal smoothing parameter for training the network. The proposed approach was evaluated on benchmark phishing datasets obtained from UCI machine learning repository and on our Phish_Net dataset. The experimental results reveal that the proposed PNN with MHS_GR (PNN_HS3) obtained 98.53%, 96.92%, and 97.12% of detection accuracy and 2.02%, 3.39%, and 3.12% of FPR for UCI_1, UCI_2, and Phish_Net dataset respectively.