Deep fake technology has been advancing rapidly and has therefore imposed major challenges to security, digital authenticity, and public confidence in the media. To tackle these issues, this research introduces a new deep fake detection framework that combines the Xception-based model for feature extraction with a temporal convolutional network (TCN) for temporal sequence analysis. Our approach improves accuracy by detecting both spatial and temporal anomalies, unlike conventional models that solely rely on spatial features. Our model outperforms state-of-the-art methods by 76.00% in the F1-score, achieving an accuracy of 85.28% when tested on benchmark datasets like FaceForensics++, DFD dataset, and Celeb-DF. The framework proposed is found to be more robust against compression artifacts and adversarial attacks, and hence a sound solution for real-world use. Therefore, this work proposes a novel deep fake detection framework that leverages the high-power Xception-based model for feature extraction and TCN for temporal sequence analysis.