Air quality prediction in Sarajevo using Machine Learning Methods

MSc student: Emina Dzaferovic

Mentor: Assist. Prof. Dr. Kanita Karadjuzovic-Hadziabdic

Due to the fact that air pollution is a complex mixture of toxic components that has the direct impact on human health, life quality and environment, prediction of it became a priority in the last couple of decades. In this study, meteorological variables and concentration of air pollutants were used to predict the common air quality index (CAQI) for an hour/day for the Bjelave station. Prediction models for CAQI were built using some of the most popular machine learning techniques used in this domain, including support vector regression (SVR), random forest (RF), decision tree (DT), decision tree with AdaBoost (DT WITH AB), principal component analysis with logistic regression (PCA WITH LR), extreme gradient boosting (XGBOOST), multiple linear regression (MLR) and multi-layer perception with three and four layers (SHALLOW and DEEP), and were trained on a three-year period data. Prediction performance was measured using regression metrics: R-squared, MSE, RMSE, MAE as well as the accuracy. Best evaluated model in majority of cases is an ensemble technique, random forest with approximate accuracy of 80% and 100% where input variables are meteorological alone and where input variables are meteorological with the concentration of air pollutants, respectively.