Intial Problem Statement

The City of Baton Rouge continues to struggle with widespread blight that affects the look, safety, and economic health of neighborhoods. Abandoned buildings, overgrown lots, illegal dumping, and neglected properties keep spreading. Blight touches every part of the city, but it hits hardest in lower-income and historically disinvested neighborhoods, lowering property values and raising safety risks. It also burdens city departments that are already stretched thin on staff and tools. Today, the city’s response is mostly reactive, driven by complaints through our 311 system, which means they often act too late to salvage a property and stay stuck playing catch-up instead of getting ahead of the problem.

This project aims to reduce the prevalence of blight and abandoned properties across the city and improve neighborhood economic health in an efficient, effective, and equitable manner. We plan to do this by using historical data about inspections, complaints, and remediations to prioritize properties in need of proactive interventions and connect them with those interventions, including commercial participation and the distribution of remediation resources.

Post-Summer Executive Summary of the Report

Baton Rouge and the unincorporated areas of East Baton Rouge Parish face numerous deteriorating properties due to population decline, a devastating flood in 2016, and a complex property-law system where most estates are inherited, making title clearance difficult. Urban blight reduces property values and the tax base, threatens public safety, and disproportionately affects marginalized neighborhoods. The Department of Development’s (DoD) response to this growing issue is largely reactive to citizen complaints, leaving unreported properties neglected and allowing cases in the system to worsen during the lengthy inspection and remediation process.

This report describes a collaboration between the Data Science for Social Good Fellowship at Johns Hopkins University and City of Baton Rouge Department of Development to develop a machine learning system to help the City proactively identify and address urban blight. We treated this as a prioritization problem. Of the roughly 200,000 parcels in the parish, the department can proactively inspect only a small number in any given cycle, so the goal was to rank the properties it has authority over by their risk of experiencing severe blight and provide inspectors with a short set of high blight risk properties to be prioritized for inspection. The eligible set is every parcel in the department’s jurisdiction that has not had a case opened on it in the prior year, about 165,000 parcels on a typical date. We mark a parcel as severely blighted when the city records a strong signal within the following year, such as an open and unsecured structure, condemnation or demolition, structural or fire damage, or a Condemnable or Emergency blight survey score, and we train only on properties the city has actually engaged with, since ones it has never visited carry no observed outcome.

Our machine learning models suggest that shifting to a proactive, predictive approach could potentially improve on the department’s current complaint-driven approach. The models show the potential to identify severely blighted properties an average of 13 months earlier than historical discovery timelines, which might buy the city critical lead time for assistive interventions. Furthermore, long-term outcome analyses indicate the models might uncover hidden, severe blight that existing tracking systems could otherwise miss. To explore how to operationalize these findings, we developed an interactive dashboard for case managers; however, a randomized field trial is a necessary next step to properly validate the tool’s true effectiveness on previously uninspected properties.