A PREDICTIVE DATA SCIENCE FRAMEWORK FOR AUDITING PUBLIC TRANSPORT EQUITY: THE MOBILITY DESERT INDEX

Urban mobility is a critical determinant of spatial equity, yet audits based only on static geographic buffers may miss operational conditions. This paper presents an integrated Data Science framework centered on the Mobility Desert Index (MDI), which combines socioeconomic pressure (P), network circuity (I), and audited supply and regularity (O). A controlled and reproducible benchmark uses 50 route-level units, 3,000 stop locations, 15,000 fleet observations, and 20,000 demand points. Isolation Forest recovered injected anomalies with precision, recall, and F1 of 0.153. In five-fold stratified evaluation, linear SVM and logistic regression obtained the highest observed mean macro-F1 of 0.744, with dispersion too large to establish superiority. Under a ten-unit integer budget, a greedy marginal rule is exactly optimal for this separable formulation, so population-based search is not warranted here. The results establish an auditable workflow for mobility-desert diagnosis, classification, and equity-oriented supply allocation.