Estimating PM2.5 Concentration Using Multiple Linear Regression (MLR) and the Random Forest (RF) Approach in Jakarta, Indonesia

Jakarta, a vast urban sprawl, undergoes rapid economic development accompanied by a vast population, land-use change expansion, and increased energy demand for transportation and industrial activities, resulting increase in PM2.5 concentrations. Efforts take place by governments to address urban air pollution by establishing air quality monitoring networks. However, the number of networks is still sparse and insufficient to record long-term series PM2.5 data. Thus, providing real-time continuous measurement of PM2.5 concentration remains challenged due to the inadequacy of ground-based measurements and expensive maintenance. Without long-term series PM2.5 data, assessing high-risk air pollution exposure is hard to quantify. This study is the first to estimate PM2.5 concentration using a combined MLR and RF in Jakarta megacity, Indonesia. The daily 24-h averaged ground-level PM2.5 concentrations are estimated by using meteorological parameters such as relative humidity (RH), visibility, 24 h average temperature (TAV), minimum temperature (TMIN), maximum temperature (TMAX), and dew point (DP) in 2016–2020. MLR and RF models were developed for dry, wet, and overall seasons. Using cross-validation (CV) with 75% training data, both MLR and RF models were revealed to agree with ground observed data. RF performed slightly better than MLR in wet, dry, and overall seasons with mean correlation (mean MAE) = 0.69 (7.55) µg m−3, 0.58 (9.40) µg m−3, and 0.72 (8.94) µg m−3, respectively. The correlation value could improve more whenever long-term data is available and include other variables as model inputs.