Data-Driven Strategies for Mobile Photo Enforcement: Leveraging Machine Learning and Inclusive Impact Evaluation for Effective Deployment

Mobile Photo Enforcement (MPE) programs have been proven effective to reduce speeding. However, the deployment of MPE still faces two critical gaps: understanding how MPE deployment impacts residents and whether such impact is fairly distributed considering demographics, and optimizing the allocation and scheduling of MPE to maximize safety benefits while ensuring distributional equity in deployment. This thesis addresses these challenges through a comprehensive MPE Distributional Analysis and the development of a machine learning-based scheduling framework. The MPE Distributional Analysis examined MPE deployment patterns, analyzing distributional equity from both procedure (enforcement presence) and outcome (ticketing distribution) perspectives. The analysis focused on three key questions: whether MPE presence correlates with neighbourhood demographics, how site characteristics influence deployment duration and ticketing levels, and whether ticket receipt rates vary by neighbourhood demographic composition. To answer these questions, descriptive analysis and regression analysis were applied. Results showed that enforcement presence was similar when residents travelled within their neighbourhoods regardless of demographic condition but varied when they travelled outside their neighbourhoods. Regression analysis results showed that MPE deployment was primarily driven by safety considerations such as traffic volume, road density, and collisions, rather than demographic factors such as ethnicity, age, or income levels. Additionally, the number of tickets per household remained statistically constant across demographic groups, indicating that the current MPE deployment did not specifically impact any group. To improve deployment efficiency while maintaining distributional equity, a machine learning-based scheduling framework was developed. The framework addressed the challenge of heterogeneous MPE data by reformulating MPE performance prediction as a classification problem. MPE performance was classified as "Ideal," "Adequate," and "Developing," based on speeding and collision conditions. Machine learning methods were applied to predict MPE performance using engineered temporal, spatial, historical patrol, and collision features as inputs. Both individual machine learning models and the automatic machine learning tool AutoGluon were tested. AutoGluon achieved the highest overall accuracy of approximately 68.5\%. Comparative analysis showed that even under conservative worst-case assumptions, ML prediction-based schedules could increase the proportion of Ideal and Adequate enforcement opportunities, achieving over 50\% improvement in high-value site coverage. Additionally, SHAP (Shapley Additive Explanations) analysis was used to rank feature importance and contribution. It revealed two critical operational insights to optimize resource allocation and improve safety benefits: increasing the interval for sites that have been recently and frequently visited, and deploying enforcement in a timely manner at sites that have recently recorded frequent collisions. The schedules can be derived by solving a mixed-integer linear programming model using ML predictions as inputs. Finally, a Gini index-based post-processing step was incorporated into the framework, enabling selection among safety-optimized schedules that can improve distributional equity in enforcement deployment across demographic groups. This thesis makes three primary contributions: (1) providing a comprehensive distributional analysis of MPE programs that examines equity in both MPE allocation and outcome, (2) offering empirical evidence demonstrating that Edmonton's current MPE deployment prioritizes safety without causing discriminatory impacts across demographic groups, and (3) developing a practical ML-based scheduling framework that improves scheduling efficiency while maintaining equity considerations. The research supports evidence-based decision-making that optimizes enforcement efficacy while maintaining fairly allocated resources across demographic groups.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.