Machine Learning Based Improved Phishing Detection using Adversarial Auto Encoder

Phishing scams are frequently used by malicious actors to get login credentials from unsuspecting users. One method for detecting phishing is machine learning. Despite the fact that the accuracy of machine learning algorithms is mostly determined by the training data, there isn't much easily accessible attack data. We present a method for enhancing the datasets used by machine learning algorithms. We create samples that resemble phishing websites using an Adversarial Auto Encoder (AAE) and offer metrics to judge the effectiveness of the samples. We compare these samples to models created from actual data. Some of the created samples could be impossible for the current model to recognize. After that, training is done with the remaining samples. Improved levels of accuracy and reliability are provided by the upgraded machine learning models. In other words, phishing detection is improved during model construction by merging data from real phishing sites with AAE fake data.