Real-Time Air Quality Forecasting in Smart Cities Using Edge–Cloud IoT Architecture and Filtered Sensor Data

Low-cost IoT sensors deployed for particulate matter (PM2.5) monitoring are often susceptible to measurement noise, outliers, and environmental interference, which can severely degrade the accuracy of datadriven forecasting models. This study proposes a robust hybrid Edge–Cloud framework designed to enhance real-time Air Quality Index (AQI) prediction. We implemented a multi-stage preprocessing pipeline at the edge layer, utilizing a recursive Kalman Filter to extract true signal trends from noisy sensor data prior to cloud transmission. Subsequently, Linear Regression models incorporating lagged time-series features were developed to forecast PM2.5 concentrations. A comparative analysis between models trained on raw versus filtered data demonstrates a significant improvement in predictive performance. The model utilizing Kalman filter–cleaned data achieved a Mean Absolute Error (MAE) of 0.34 and a Root Mean Squared Error (RMSE) of 0.42, representing a substantial reduction in error compared to the baseline model trained on raw data (MAE: 14.24, RMSE: 16.87). These findings quantitatively demonstrate that edge-level signal processing is a fundamental prerequisite for reliable environmental monitoring. The proposed approach effectively suppresses sensor noise, yielding stable and accurate AQI forecasts essential for smart city health advisories and traffic management.

Hali tarjima qilinmagan