AI-enabled Air Quality Predictive Model for Fine Particulate Matter Incorporating Meteorological Data: A Case of Dar es Salaam City

Air pollution caused by particulate matter (PM2.5 and PM10) poses major environmental and public health challenges in rapidly urbanising cities such as Dar es Salaam, Tanzania. However, effective air quality monitoring in developing countries remains limited due to the high cost of regulatory-grade systems and the susceptibility of low-cost sensors to calibration drift, environmental interference, and measurement noise. In addition, limited integration of meteorological variables and calibration correction mechanisms reduces prediction reliability and long-term monitoring performance. This study proposes an AI-enabled supervised machine learning model for air quality prediction using calibrated low-cost sensor data integrated with meteorological information. PM2.5, PM10, temperature, and relative humidity data were collected from seven monitoring stations across Dar es Salaam between April and May 2026 using PurpleAir sensors. Calibration drift correction was performed using reference measurements from the Tanzania Meteorological Agency (TMA). The datasets were used to develop Multiple Linear Regression (MLR), Polynomial Regression (PR), Decision Tree Regression (DTR), and Random Forest Regression (RFR) models for Air Quality Index (AQI) prediction. Model performance was evaluated using the Coefficient of Determination (R2), Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE). The results show that integrating meteorological variables and calibration correction significantly improves prediction performance and reliability. Polynomial Regression achieved the highest R2 value of 0.9815, Random Forest Regression achieved the lowest RMSE value of 0.3956, and Decision Tree Regression achieved the lowest MAE value of 0.0103. Overall, Random Forest Regression provided the most balanced performance across all evaluation metrics. The findings further reveal that temperature, relative humidity, and seasonal variability substantially influence particulate matter concentrations and AQI prediction performance, while calibration correction reduced systematic sensor bias and improved measurement consistency. The proposed model provides a scalable, cost-effective, and intelligent solution for real-time air quality monitoring and prediction. The study contributes to the development of reliable AI-based environmental monitoring systems for public health protection, evidence-based policymaking, climate resilience, and smart city environmental management in developing regions.