An Effective Neural Network Phishing Detection Model Based on Optimal Feature Selection

As a common means to obtain user privacy information, phishing poses a big threat to people's daily network environment. The detection and prevention the threats of phishing websites are of importance. Due to the active learning ability from large-scale datasets, neural network is an important heuristic machine learning method in phishing websites detection and prevention. However, during the process of data training, some useless features may cause the machine learning method to over-fitting which will result in the training model not being able to precisely predict and detect the phishing websites. Aiming at this problem, based on the optimal feature selection method, this paper proposes an effective neural network detection model (OFS-NN) to detect the pushing websites. Under this model, an optimal feature selection algorithm that adapts to the sensitive features of phishing URLs (Uniform Resource Locators) is firstly proposed. Based on the calculation of the effective value of each feature, this algorithm sets a threshold to eliminate some useless features and selects an optimal feature set suitable for detecting phishing websites. Then, the selected optimal feature set is trained by the neural network to construct an optimal classifier to classify and predict the pushing websites. The experimental results have shown that the proposed OFS-NN provides an effective solution for predicting and detecting phishing websites. It has little false positive rate and strong generalization ability. In addition, the optimal feature selection algorithm improves the performance in the sample training process of machine learning methods.