Support Vector Regression for PM10 Concentration Modeling in Santa Marta Urban Area

This paper presents a method for particulate material PM10 modeling based on support vector regression (SVR). Specifically, we applied epsilon-support vector regression (epsilon-SVR) and nu-support vector regression (nu-SVR) to a set of data recorded in the city of Santa Marta, Colombia, between 1999 and 2016. The set of data was initially pre-processed, filtered and normalized, and then was used to fit the SVR models. The parametrization and accuracy of each regression model are reported here. We used a month as the unit of time for the models and analyzed the accuracy for one-step predictions. The final results of this work show the best parameters and prediction properties of the SVR models for pollution data modeling in Santa Marta.