COMPARISON OF XGBOOST AND RANDOM FOREST METHODS IN PREDICTING AIR POLLUTION LEVELS

Abstract

Air is one of the elements needed by living things, including humans, to survive. The air quality in an area also affects the health and quality of human life and its surrounding environment. However, with the current phenomenon, the influence of the increasing number and mobility of humans actually degrades air quality, caused by the pollutants produced. For further impacts, poor air quality can reduce human life expectancy. Big cities in Indonesia, such as Surabaya, also experience the same thing due to the lack of public awareness of air pollution. The biggest contributors to air pollution are motor vehicles and industrial activities that emit carbon monoxide (CO), nitrogen oxides (NO), ozone (O3), and other particles (PM10). This condition is addressed by the Surabaya City Government by installing air condition measuring devices at points considered prone to pollution. This device works to measure urban air conditions daily and provides data that can be utilized to establish strategic policies. By utilizing the data, in this research, we implemented two prediction methods from machine learning technology, namely XG Boost and Random Forest. In accordance with the objective of this research, both methods will be compared for accuracy in predicting air pollution levels in Surabaya based on Ozon (O3) substance within the period of January 1, 2020, to December 31, 2020. Both of them have a similarity in that they implement tree-ensemble based, which are appropriate for handling non-linear data. The XG Boost method managed to achieve the best error value of 0.0510, and the Random Forest method reached the best error value of 0.0468.