Illegal parking in urban environments disrupts traffic flow, causes greenhouse gas emissions, and poses a threat to pedestrians and cyclists. Traditional Intelligent Transportation Systems (ITS) are based on high-cost surveillance and video analysis that typically does not take into account the dynamics and complexity of the urban environment. To address these limitations, this study overcomes this gap by proposing an intelligent parking violation prediction framework using a hybrid Spatio-temporal Graph Neural Network (STGNN) approach, which combines Graph Kolmogorov-Arnold Networks (GKAN) and Liquid Neural Networks (LNN). The GKAN model excels at uncovering intricate spatial patterns in the urban dataset, while the LNN model has intrinsic dynamic temporal variations in real time due to its adaptive learning capability. This integration of spatio-temporal relationships of metropolitan datasets is effective modeling and thus can be robust across diverse urban scenarios. The proposed approach is well-suited for practical usage in real-world applications and achieves a high prediction accuracy with a high \(R^2\) score of 0.95, and shows significant improvement in other metrics such as MAE and MSE. These results underscore the performance of the proposed GKAN-LNN framework in addressing the challenges presented by the parking violation prediction task to develop safe, sustainable, and well-governed urban settings.