Gait recognition plays a vital role in healthcare, security, and smart environments. Traditional deep learning models, including Capsule Networks (CapsNets), often struggle to capture the spatial-temporal dependencies in inertial sensor data fully. In this paper, enhanced GCN-TCN-CapsNet is a novel hybrid architecture that integrates Graph Convolutional Networks (GCNs) for adaptive spatial feature extraction, Temporal Convolutional Networks (TCNs) for efficient local temporal modeling, and Capsule Networks (CapsNets) for hierarchical representation learning. Unlike prior approaches, our model incorporates dynamic edge creation using the k-nearest neighbors (KNN) method, enabling adaptive inter-sensor relationship modeling. Extensive experiments on the four WHU-Gait person identification datasets demonstrate that the proposed model consistently achieves validation accuracies exceeding 97%, highlighting its effectiveness in gait recognition tasks; this model can work with real-time data, making it suitable for continuous and non-invasive authentication in security and medical applications.