Truth Revealed: Enhancing Deception Detection Using Long-Term Recurrent Convolutional Networks

Identifying when someone is lying is a challenging but critical task, especially in fields like law enforcement and security, where detecting deception can have significant consequences. This research paper explores the development of deep learning techniques for deception detection using Long-Term Recurrent Convolutional Networks (LRCN) applied to the Dolos dataset. Different LRCN models were developed and evaluated on various performance metrics for detecting deception using visual features from preprocessed data. The evaluation involved hyperparameter tuning and feature selection techniques on manually annotated features within the MUMIN scheme of the Dolos dataset, achieving remarkable results, with 72.88 % accuracy. These results surpassed the previous state-of-the-art accuracy of 66.84 %. Additionally, a Graphical User Interface (GUI) was developed to apply the model in real-world settings, marking a significant advancement in deception detection with promising future applications.