Advance Deception Detection using Multi-Modal Analysis
Krisha Patel, Priyanshi Airen, Swayam Singh · 2025
Humans have consistently failed to convincingly cheat detection, relying in the past upon intuition or polygraph, both being fundamentally faulty in reliability. Advances in artificial intelligence (AI), machine learning, and computer vision have made it possible for more effective and efficient deception detection systems. This paper provides a cutting-edge multimodal deception detection system that integrates text, video, image, and audio analysis to identify untruthful behavior effectively. The proposed real-time system employs Bidirectional LSTM networks for text processing, vocal feature extraction using TensorFlow-Based models, and real-time vision pipelines with OpenCV to detect visual deception indicators such as microexpressions and eye movements. Early multimodal fusion is conducted in the process of data management, increasing synchronization and accuracy over the traditional late fusion techniques. Experimented with against our own implementation on the Dolos and PolitiFact datasets, our model registered substantial performance metrics of precision (85. 12%), recall (82. 12%), and F1 score (83. 98%), indicating it can differentiate between deceitful and genuine actions strongly. The model further has dynamic thresholds to enable greater sensitivity to inconclusive cases with some more tuning to be achieved. Our solution despite data limitation challenges and computation cost remains an important milestone to enabling accurate real-time multimodal deception detection that can be applied in various industries including law enforcement, human resource management, and security.