Bayesian Optimization based Hyperparameter Tuning of Ensemble Regression Models in Smart City Air Quality Monitoring Data Analytics

This paper uses the Bayesian optimization for fitting Ensemble regression models for tuning the machine learning model hyperparameters with reduced computation. We use the Pune Smart City air quality monitoring dataset with temporal variation of hazardous chemical pollutants in the air. The aim here is to reliably predict the suspended particulates as the air quality metrics using other environmental variables, considering linear models and nonlinear ensemble of tree models. To achieve good predictive accuracy a computationally expensive optimization method is required which has been achieved using the Gaussian Process surrogate assisted Bayesian optimization. We also show the diagnostics plots of the residuals from the nonlinear models to explain model quality.