Predicting demand for 311 non-emergency municipal services: An adaptive space-time kernel approach

Abstract Many cities in the United States and Canada offer a 311 helpline to their residents for submitting requests for non-emergency municipal services. By dialing 311, urban residents can report a range of public issues that require governmental attention, including potholes, graffito, sanitation complaints, and tree debris. The demand for these municipal services fluctuates greatly with time and location, which poses multiple challenges to effective deployment of limited resources. To address these challenges, this study uses a locally adaptive space-time kernel approach to model 311 requests as an inhomogeneous Poisson process and presents an analytical framework to generate predictions of 311 demand in space and time. The predictions can be used to optimally allocate resources and staff, reduce response time, and allow long-term dynamic planning. We use a bivariate spatial kernel to identify the spatial structure and weigh each kernel by corresponding past observations to capture the temporal dynamics. Short-term serial dependency and weekly temporality are modeled through the temporal weights, which are adaptive to local community areas. We also transform the computation-intensive parameter estimation procedure to a low dimensional optimization problem by fitting to the autocorrelation function of historical requests. The presented method is demonstrated and validated with sanitation service requests in Chicago. The results indicate that it performs better than common industry practice and conventional spatial models with a comparable computational cost.