Predictive models for dissolved oxygen in an urban lake by regression analysis and artificial neural network

Abstract

This paper portrays predictive models for Dissolved Oxygen (DO) levels in an urban lake using common water quality parameters like Temperature, pH, Conductivity and Oxidation Reduction Potential (ORP). Data were sampled using three real-time industry-standard sensors those are OPTOD, CTZN, and PHEHT, and then interpolated using the ArcGIS interpolation technique. Correlation studies were analyzed through the Machine Learning (ML) algorithm, the correlation study signified a positive linear correlation with DO against pH, temperature, salinity and conductivity and the model was corroborated by R-score which came to 0.687 and RMSE was 0.834. Multiple Linear Regression (MLR) model was developed to predict the DO with the correlated data of water parameters. In addition, an Artificial Neural Network (ANN) method using the Levenberg-Marquardt algorithm was developed to build a model to predict the DO as well. Then, the models’ performance was validated and the R² accuracies were 0.963 for MLR and 0.93 for ANN and models were checked for the predicted data against the actual data. The appropriateness of the ANN model for forecasting investigated attributes is indicated by the fact that the discrepancy between the forecasted and real ANN model is significantly lesser than that of the regression model. The developed equation in this paper can be used to reveal DO data from unknown urban lake water.