PrEACT: Predictive Learning for Early Action in Civil Complaints with AI and Temporal Modeling

This paper presents a novel approach for predicting citizen civil complaints related to public safety—such as potholes, illegal dumping, and unauthorized installations on roads—using a hybrid deep learning model. The proposed model combines Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks. We utilize an unprecedented nationwide dataset comprising three years of civil complaints collected across all regions of South Korea (2021–2023), aggregated into 1 km spatiotemporal grid cells to ensure privacy while maintaining spatial resolution for analysis. These spatiotemporal cells enable AI-driven predictive analytics that support proactive urban management. Our framework, Predictive Learning for Early Action in Civil Complaints with AI and Temporal modeling (PrEACT), captures both spatial distribution and temporal trends of complaints, providing predictive insights into when, where, how, and what issues are likely to arise. Experimental results demonstrate that our hybrid model outperforms both a baseline approach and a Random Forest model in terms of Root Mean Square Error (RMSE). In addition, prediction results are visualized through interactive civil complaint maps, making insights easily accessible to both the public and government agencies. To the best of the authors’ knowledge, this is the first attempt to apply civil complaint data in a microscale, grid-based manner across both urban and rural contexts. This work underscores the importance of integrating AI and big data into public service delivery and lays the foundation for more proactive, data-driven policy interventions.