Estimating the Water Quality Class of a Major Irrigation Canal in Odisha, India: A Supervised Machine Learning Approach

Abstract

Contamination of surface water by rapid industrialization, natural and anthropogenic activities is of great concern over the last few decades. Nowadays, canal water systems are no exception to this form of contamination, which results in water quality degradation. To classify the canal water based on the Bureau of Indian Standards (BIS), it was thought to develop a quick and inexpensive approach as an alternative to the time-consuming analysis approach. With this motivation, the present study explores building a machine learning model for water quality classification of a major canal namely the Talaldanda canal operating in the state of Odisha, India. The water quality class is predicted using supervised machine learning (ML) prediction models for the new canal water input parameters. The water quality parameters such as pH, dissolved oxygen (DO), biochemical oxygen demand (BOD), and total coliform (TC) at six strategic locations of the canal from the year 2013-2020 were collected from Odisha State Pollution Control Board for the training phase. The supervised ML models used in the study are Decision Tree (DT), Neural Network (NN), k-NN (k-Nearest Neighbor), Naïve Bayes (NV), Support Vector Machine (SVM), and Random Forest (RF). The predictions of the models are evaluated using the Orange-3.29.3 data analytics tool. When analyzing the performance parameters by sampling the training data into training and testing using cross-validation, the results show that DT has a higher classification accuracy (CA) of 96.6 percent than other ML models. In addition, the likelihood of DT correctly predicting water quality class for the testing dataset is higher than that of other prediction models.

Keywords

• taladanda canal, water quality class, supervised machine learning, prediction model, classification accuracy