Prediction of biochemical oxygen demand in mexican surface waters using machine learning

Abstract

The monitoring of surface water quality is insufficient in Mexico due to the limited water monitoring stations. The main monitoring parameter to evaluate surface water quality is the biochemical oxygen demand. This parameter estimates the biodegradable organic matter present in the water. Concentrations above 30 mg/l indicates a high level of contamination by domestic and industrial waste. Therefore, the aim of this work to provide a reference to the conventional process of determining biochemical oxygen demand using machine learning. The database used was collected by the National Water Commission (CONAGUA). Pearson’s correlation and Forward Selection techniques were applied to identify the parameters with the most important contribution to prediction of biochemical oxygen demand. Two groups were formed and used as input to four machine learning algorithms. Random forest algorithm obtained the best performance. Group 1 and 2 of parameters obtained a 0.76 and 0.75 coefficient of determination respectively. This allows choosing an adequate group of parameters that can be determined with the chemical analysis instruments available in the study area.