Smart city initiatives are being promoted globally to address significant security challenges and build confidence in the transformative potential of digital technologies. Human-Centric Artificial Intelligence (HCAI) focuses on a process perspective design that augments human capabilities, supports equity, and upholds ethical principles. This study proposes a model based on a Hybrid Spatial Weighted Attention Layer Pooling Convolutional NeuroNet by incorporating human feedback to detect anomalies in smart cities using a CCTV violence detection dataset. The model incorporates predictions and visualization utilizing HCAI-based Local Interpretable Model-agnostic Explanations (LIME). The preprocessing phase involves normalization to standardize image input and ensure uniformity, while data augmentation techniques, such as rotations, flips, scaling, and cropping, are applied. The proposed LSTM-CNN-based model demonstrates an accuracy of 98.25%, precision of 98.26%, recall of 98.25%, and an F1-score of 98.24%. Furthermore, future study directions are highlighted. This study represents a significant advancement in the accuracy and reliability of surveillance systems, improving public safety and monitoring efficiency in smart cities.