The paper introduces a robust deep-learning technique, ResNet-Swish-BiLSTM, for identifying fraudulent deepfake videos. The method leverages the Swish activation function, which combines the benefits of ReLU and Sigmoid, to enhance the model's learning behavior and feature extraction capabilities. The proposed model is evaluated using the Face Forensics++ (FF++) and Deepfake Detection Challenge (DFDC) datasets, achieving 96.23% accuracy on FF++ and 78.33% on aggregated records from both datasets. The model's effectiveness is demonstrated through extensive experiments, showing superior performance compared to existing deepfake detection techniques. The study also highlights the model's resilience to various visual manipulations and its ability to handle temporal variations in deepfake content. The authors conclude that the ResNet-Swish-BiLSTM model is a promising approach for advanced digital forensics and criminal investigations.The paper introduces a robust deep-learning technique, ResNet-Swish-BiLSTM, for identifying fraudulent deepfake videos. The method leverages the Swish activation function, which combines the benefits of ReLU and Sigmoid, to enhance the model's learning behavior and feature extraction capabilities. The proposed model is evaluated using the Face Forensics++ (FF++) and Deepfake Detection Challenge (DFDC) datasets, achieving 96.23% accuracy on FF++ and 78.33% on aggregated records from both datasets. The model's effectiveness is demonstrated through extensive experiments, showing superior performance compared to existing deepfake detection techniques. The study also highlights the model's resilience to various visual manipulations and its ability to handle temporal variations in deepfake content. The authors conclude that the ResNet-Swish-BiLSTM model is a promising approach for advanced digital forensics and criminal investigations.