PSO active learning of XGBoost and spatiotemporal data for PM2.5 sensor calibration

Abstract

Ambient PM2.5 concentrations affect human health and natural environment. Government-built PM2.5 monitoring supersites are accurate but cannot provide a dense coverage of the air quality index (AQI) monitoring. Broadly-distributed PM2.5 microsite sensors can complement supersites for fine-grained monitoring. However, due to the low cost of microsite sensors, the accuracy of the raw AQI measurements is not high enough for monitoring purpose. Calibration of low-cost sensors is thus a necessary processing step to enhance measurement fidelity. This paper presents a particle swarm optimization (PSO) based active learning of optimal configurations of XGBoost and spatiotemporal data for PM2.5 microsite sensor calibration. The experimental results show that PSO active learning of the optimal configurations of XGBoost and spatiotemporal data can calibrate low-cost PM2.5 microsite sensors to achieve high accuracy by reference to governmental supersites.