A Binary Classification Model for PM10 Levels

The particulate matter in the air effects human health in a negative way. Yet, no regression model has estimated the density of PM10 at Istanbul using datasets with imbalanced class distribution. In order to MI this gap, we designed a new regression model that transforms the regression problem into the imbalanced binary classification problem at the initial stage. In this paper, PM10 classification problem is considered as the imbalanced binary classification problem that is coded as harmless class (1) and dangerous class (0). In the sampling part of the solution, the balanced version of the data by Under Sampling methods yielded unsatisfactory results. In the algorithmic part, the performances of RFC (Random Forest Classifier), ETC (Extra Trees Classifier) and GBC (Gradient Boosting Classifier) models, which stand out with their positive effects on unbalanced learning problems, are compared in terms of AUROC. The proposed model, uses all training set samples and predicts through RFC. The experimental results on real world dataset seem quite promising for our further research.