Explainability-Aware Trust-Driven Hybrid Machine Learning Framework for Transparent Smart City Decision Systems Using Open Urban Data

Modern societies are being rapidly urbanized, which has led to an increasing reliance on datadriven technologies that seek to streamline various functions of cities such as transportation, environmental surveillance, energy conservation as well as city security. However, the current state of machine-learning (ML) models is that a significant number of these models are blackbox systems that hinder openness, confidence, and responsibility in the governance of the people. To address these concerns, in this paper, we have proposed a hybrid framework of ML with trust and explainability guarantees that can provide clear decision support towards smart-city applications using publicly available city data. The model proposed will consist of a Random Forest (RF) classifier combined with an XGBoost classifier risk into an ensemble architecture, but at the same time, will compose posthoc mechanisms of explanation, namely SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explainations (LIME) to produce full world and local explanations of model predictions. In addition to the proposed factor of generation of explanations, the framework also has a special Trust Evaluation Module, which quantifies the reliability of such explanations in terms of three rigorously defined measures, namely: (1) fidelity which quantifies the consistency of explanations and the underlying model behavior to input perturbations; (2) stability which quantifies the consistency of prediction and explanation to input perturbations or crossvalidation folds (3) consistency which quantifies the degree to which different explanations methodologies or across cross-validation folds are consistent. Empirical analyses that were conducted using real-world smart-city datasets, such as air-quality surveillance imagery and urban stream of IoT sensor data, provoke the idea that the hybrid model shows higher predictive integrity compared to baseline models and this approach earns a higher score in terms of trust, which promotes greater transparency and accountability in the use of AI-enabled urban governance.