Optimizing Deep Neural Networks Using ANOVA for Web Phishing Detection

Abstract

Phishing attacks are crimes committed by sending spoofed Web URLs that appear to come from a legitimate organization in order to obtain another party's sensitive information, such as usernames, passwords, and other confidential data. The stolen information is then used to commit fraud, such as identity theft and financial fraud, and can cause reputational damage to the party that is the victim of the phishing attack. This can cause great harm to the victimized individual or organization. To overcome these problems, this research uses feature selection using ANOVA and Deep Neural Networks (DNN) to detect web phishing attacks. Feature selection is used to optimize the performance of the DNN model to achieve more accurate results. Based on the results of feature selection using ANOVA, there are 52 attributes that have a significant impact on web phishing attack detection. The next step is to implement DNN to build a web phishing attack detection model. The results of testing the web phishing detection model show that in the training phase, the accuracy value increased by 17.51% for the 80:20 dataset and 18.39% for the 70:30 dataset. During the testing phase, the accuracy value increased by 17.8% for the 80:20 dataset and 18.58% for the 70:30 dataset. The resulting recognition model shows consistent and reliable results not only during training, but also during testing in situations closer to real-world conditions. Conclusively, the use of ANOVA proves effective in mitigating less relevant features and contributing to the optimization of web phishing detection models.

Keywords

• web phishing detection
• anova
• deep neural networks
• feature selection
• optimizing