Development of a Dual CNN-LSTM Model for The Detection of Deepfake Video Cyber Phishing Attacks

Deepfake technology has introduced new complexities into cyber-phishing attacks by enabling cybercriminals to create authentic-looking and artificial intelligence altered videos to impersonate trusted individuals. These synthetic videos can deceive users more effectively than traditional phishing techniques, which rely on text or email phishing attacks. As such, existing cybersecurity mechanisms, primarily designed to detect textual or URL-based threats, fall far short when it comes to video-based deception detection. This research developed a dual CNN-LSTM model for the detection of deepfake video cyber phishing attacks. To achieve this, a dataset of 5,000 labeled videos were collected from the Cyber Institute of Atlanta, USA. The ResNeXt CNN were applied to extract the features, capturing spatial patterns within the extracted frames. The dataset was splitted into 70% training and 30% test sets. The model was trained using a merged benchmark dataset of both the Deepfake Detection Challenge and the FaceForensics++ and a dropout rate of 0.4 was applied to mitigate overfitting. The experimental results showed that the model had 99.81% detection accuracy, indicating a high overall correctness in its predictions. The precision score of 99.73% reflects the model’s strong ability to correctly identify deepfake phishing content without misclassifying genuine videos. The recall rate of 99.87% confirms that the model effectively detects the majority of deepfake instances, minimizing false negatives. The combined performance is further substantiated by an F1-Score of 99.79%, which balances the trade-off between precision and recall, showcasing the robustness of the model. The results demonstrated that the proposed model had high reliability when differentiating between real and altered video content. The model is therefore recommended for the detection of deepfake videos to prevent phishing attacks in a computer network environment. Keywords: Deepfake Technology, Security, Phishing Attacks, Artificial Intelligence, Convolutional Neural Networks, Long Short-Term Memory. Aims Research Journal Reference Format: Akinwumi D.A., Orojo, T. E., Akingbesote. B.O. & Aliyu, E. O. (2025): Development of a Dual CNN-LSTM Model for The Detection of Deepfake Video Cyber Phishing Attacks. Advances in Multidisciplinary Research Journal. Vol. 11 No. 2, Pp 57-67 www.isteams.net/aimsjournal. dx.doi.org/10.22624/AIMS/V11N2P5