Neural Network based Phishing Website Detection with Using Firefly Algorithm

In the digital world, phishing is one of the cybersecurity threats that can cause security and financial problems for operators and internet service providers. Until nowadays, various techniques and methods have been proposed to detect phishing websites. In this paper, a modified multilayer perceptron neural network with Firefly meta-heuristic algorithm is used to increase the accuracy of phishing website detection. The phishing databases are taken from the UCI dataset collection. The Phishing Tank dataset under study consists of 31 features in 31 columns and 11055 rows. For missing data values, 3-nearest neighbors method was applied for predicting. In order to reduce the dimensions of the data, the principal component analysis (PCA) method was used. In this situation, 27 out of 31 features were identified as main features. For this purpose, each firefly considered as a candidate which has two dimensions. The first dimension specifies the weight of the neural network links and the second dimension determines their bias. Using firefly algorithm helps multi-layer perceptron neural network to get more exploration in the search space so it improves the probability to achieve optimal point in problem under studied. The results showed that by using the proposed method, the accuracy of phishing website detection increased by 98.2%. In other words, this method demonstrated that using a combination of multilayer perceptron neural network and firefly metaheuristic algorithm can be an effective solution for detecting phishing websites.