Graph-Based Causal Machine Learning for Self-Adaptive Control under Context Shifts in Smart Urban Mobility

Urban mobility services based on pickup–delivery routes operate under changing conditions: traffic, corridor constraints, schedules, and demand fluctuations continuously alter the context in which control decisions are executed. In many deployments, these decisions rely on fixed policies and prediction models trained on a “typical” context, which lose accuracy and effectiveness when the situation deviates from that scenario. This paper addresses how to design control mechanisms that keep the temporal performance of the service robust to context shifts, allowing the management software to adapt at runtime without manual tuning and without relying exclusively on correlational patterns. We propose a self-adaptive control approach guided by graph-based causal machine learning. The system is represented through causal graphs specific to each operational unit, instantiated here at the route level for a public school transport service, linking context variables, control decisions, and temporal outcomes. On these graphs, we train predictive models using Causally Adaptive Constraint Minimization (CACM) and integrate the resulting model into a Monitor, Analyze, Plan, Execute, and Knowledge (MAPE-K) loop that selects, according to the estimated situation, from a bounded set of timing adaptations implemented via Context-Oriented Programming (COP). The approach is validated in a simulation environment calibrated with historical records from the school transport platform, with 25,600 trips distributed across five contexts that combine demand level and route configuration. Results show that CACM maintains predictive performance comparable to a standard Empirical Risk Minimization (ERM) baseline while reducing delay variability and headway irregularity when used to drive adaptive policies. The contribution is an end-to-end architecture that connects causal discovery, robust causal prediction, and lightweight actuation for self-adaptive control in smart urban mobility.