|
| Focus | Study | Research design | Major findings |
|
| Web | Mao et al. (2017) | Collected 9,307 verified phishing websites from PhishTank as an experiment sample set. It consists of phishing pages targeting popular website (e.g., PayPal, eBay, Apple). | Phishing-Alarm, phishing attack detection solution extracts CSS-based page features, evaluates the similarity between whitelisted web pages and suspicious web page, and focuses on visual features that are hard to be tampered. |
| This study presents techniques to identify effective CSS features as well as efficient algorithms for page similarity analysis. Authors prototyped Phishing-Alarm as an extension to the Google Chrome browser and evaluated it using a wild phishing web pages. |
| Corona et al. (2017) | Empirically evaluated it on more than 5,500 web pages from compromised websites in the wild. | DeltaPhish detects compromised phishing web page by highlighting HTML code and visual difference with respect to legitimate pages hosted within a compromised website. Web pages collected in the wild from infected websites were evaluated and performed capability of detecting more than 99% of phishing web pages, while less than 1% of false detection of legitimate pages. |
| Adebowale et al. (2018) | Dataset consisted of 4,898 phishing websites, 1,945 suspicious sites, and 6,157 legitimate websites from 2 prior studies (Rami et al., 2015a, 2015b) and PhishTank. | Presents an Adaptive Neuro-Fuzzy Inference System (ANFIS) using integrated features of the text, images, and frames. This study utilized three different conventional classification algorithms (SVM, K-NN, and ANFIS). ANFIS algorithm achieved accuracies of 98.3% on web-phishing detection. |
| Abdelnabi et al. (2020) | VisualPhishNet examined 155 trusted phishing websites, which consists of 9,363 pages. | VisualPhishNet, a similarity-based detection model based on triplet convolutional neural networks (CNN), examined VisualPhish. |
|
