The increasing prevalence of deepfake videos poses significant threats to information integrity, political stability, and public trust. This study presents a dual-faceted approach: (1) developing a machine learning model for detecting deepfake videos using visual features extracted from benchmark datasets, and (2) conducting a trend analysis of deepfake content dissemination on social media platforms such as YouTube and Twitter (now known as X). Conducted using the FaceForensics++ dataset and metadata from over 2,000 social media posts collected between 2018 and 2024, this study used a fine-tuned Xception model and natural language techniques. Key findings indicate a post-2020 surge in politically motivated deepfakes and platform-specific propagation patterns. It is recommended that stakeholders implement real-time detection and awareness tools to mitigate social impact