Forensic Analysis and Detection of Spoofing Based Email Attack Using Memory Forensics and Machine Learning

Deep Learning-Based Speech and Vision Synthesis to Improve Phishing Attack Detection through a Multi-layer Adaptive Framework

“…In this research, we proposed a multi-layer adaptive framework that uses the computer vision capability of Optical Character Recognition (OCR) to read images on live phishing sites to text, and synthesize speech from uploaded deep-fake videos, while using Random Forest, and LSTM network, along with web scrapped text at various predictions layered of the framework to significantly improve the detection rate and performance of AI-based models for phishing detection. Considering the fact that existing AI-based phishing detection techniques, frameworks, and approaches can only detect text-based [32], [33], [2], [28] or URL-based phishing [27], [32], [34], [35] sites which leads to their vulnerability and inability to detect image-based, or video-based phishing sites, the proposed framework is able to overcome limitations in existing approaches, significantly improve phishing attack detection, and successfully detect complex phishing webpages with multi-dimentional deep-fake videos, images, and texts.…”

“…The dataset was split into two such that 80% was used for training, while the remaining 20% was used for validation tests. We choose random forest because of its suitability for URL-based phishing detection relative to other classifiers [26], [27], [28], [29], [30], [31].During iteration, we set both the depth and random variable to several values for optimal result but only observed a small but negligible change in the variation of the accuracy until 39. with depth > 39, the accuracy remains constant, at least till when we increase the randomness of the tree to 1 before observing little change. We finally settled on setting the randomness state to 0 so that each tree remains the same each time it is generated.…”

“…To stretch and validate our multilayered adaptive framework for its effectiveness in the detection of phishing sites containing any of (Text, videos, and images), or a combination of any, or all of the 3. It is worth remembering that all existing AI or machine learning-based phishing detection techniques and frameworks can only detect text-based [32], [33], [2], [28] or URL-based [27], [32], [34], [35] phishing sites leading to their vulnerabilities to; -phishing sites with friendly URL -phishing site on hacked legitimate domain name server (DNS) -Image-only phishing site -video-only phishing site or, combination of any of them in any order. To validate both the effectiveness and adaptability of our proposed framework in overcoming such limitations, we created 4 categories of phishing sites and uploaded them to a secure server with a compromised DNS on a friendly URL; the first was a text-only phishing site, image-only phishing site, video-only phishing site, and a phishing site combining all the features.…”