A data-driven deep learning algorithm designed to detect deceptive walking behaviour based on gait analysis. Our study involved a comprehensive user experiment with numerous participants performing both deceptive and normal walking tasks. Different subjects who walk normally and deceptively are used for training and testing. As a pre-processing step, the participants’ walking gaits as a series of 2D skeletons are extracted. By transforming the original skeleton sequences from our dataset into images, our proposed method enables the extraction of spatial and temporal features for deep network learning. Analyzed the RGB image representations to classify gait as either normal or deceptive, utilizing a convolutional neural network built from scratch, a pre-trained transfer learning approach, a long short-term memory network, and a graph convolutional network. Our analysis led us to employ pre-trained transfer learning CNN models to extract features from these images for gait classification, along with GCN for 2D skeleton classification. Experimental data demonstrate the effectiveness of our proposed strategy, highlighting its superiority over other approaches. Categorize gait walks from videos using this proposed method, making a low-cost system for the early detection of false walking possible for medical interventions.