Predictive Sensor for Biological Oxygen Demand in water using Active Learning based Random Forest Algorithm

Abstract

The environmental pollution caused by human beings, and industries is interfering with water sources thatbound the living zones in marine environments. Greywater reuse after treatment can be carried out for non-consumable water use and hence Biochemical oxygen demand (BOD) is a vital parameter in deciding the nature of waste water created. The measurement of BOD involves difficulties and performing the BOD test takes too long and the result does not remain relevant for the current wastewater. This work proposes an ensemble learning-based random forest model with active learning to iteratively select samples with large ambiguity for reducing errors in predicting BOD. The model uses minimal basic physical & chemical parameters of water and hence does not require specialized ion-selective and costly sensors for measurement. Feature importance is exploited to enhance the efficiency of the machine learning model leading to a prediction accuracy of 81.21%. The developed model could be used as a predictive sensor at the edge of the water quality monitoring system