International Conference on Computer and Knowledge Engineering

Home / 15th International Conference on Computer and Knowledge Engineering

AIRSPAN-X: Federated XGBoost with Sequential Anomaly Detection for Explainable Urban Air Quality Prediction

Authors :

Saghar Shafaati¹ S. Hossein Erfani²

1- Department of Computer Engineering, South Tehran Branch, Islamic Azad University, Tehran, Iran 2- Department of Computer Engineering, South Tehran Branch, Islamic Azad University, Tehran, Iran

Keywords :

Air Quality Prediction،Federated Learning،XGBoost،SHAP Explainability،PrefixSpan،Internet of Things (IoT)

Abstract :

Urban air quality monitoring has been of pivotal importance with its direct health impact on public welfare, especially in areas of high population density that are prone to pollution-related diseases. Conventional sensing infrastructures, although accurate, are spatially coarse and have limited capacity to identify localized pollution episodes. The current paper introduces AIRSPAN-X, a new and interpretable framework that combines adaptive federated learning, SHAP-based explainability, and sequential anomaly detection to achieve accurate and privacy-preserving air quality prediction. The new model leverages XGBoost under a federated learning framework to guarantee data privacy without compromising prediction precision over decentralized IoT sensor systems. A data-heterogeneous adaptive aggregation layer ensures robustness against data noise and non-IID distributions. The proposed framework utilizes SHAP (SHapley Additive exPlanations), a global and local explanation method, to assist stakeholders in interpreting pollutant contribution to the Air Quality Index (AQI). For identifying irregularities, PrefixSpan, a sequential pattern mining method, is used to discover temporal outliers in pollution patterns without labeled data. Experimental evaluations on real outdoor urban air quality data from 22 districts across Tehran confirm the efficacy of the framework in aspects of accuracy (up to 98.64%), scalability, and explanatory power. The proposed strategy addresses capital challenges of privacy, transparency, and irregularity detection and presents a one-stop solution to environmental monitoring in intelligent cities.