A Comparison of Regression Models for Dissolved Oxygen Prediction in Koh Yor, Thailand

Abstract

Dissolved oxygen is an indicator of the health of a water source and aquatic life. In Koh Yor, fishermen encountered the problem of sea bass dying from changes in dissolved oxygen and did not have information on dissolved oxygen levels in the water. In this paper, a system and methodology for dissolved oxygen prediction in Koh Yor water, Thailand, for breeding sea bass in cages are introduced. The major objective of this study is to find the optimal regression models that achieve the best prediction accuracy, where soft sensor information including temperature, pH, and salinity data are used. The regression model is suitable for predicting DO in open systems with non-linear water quality data. The regression models for evaluation and comparison include Gaussian Process Regression (GPR), Medium Gaussian SVM (SVM), Least Squares Regression (LSR), and Medium Neural Network (MNN), respectively. The performances of the regression models are validated with the water data set of each village collected from Koh Yor, Thailand. Experimental results demonstrate that the GPR model provides the best prediction accuracy of 91.8%, where the prediction accuracy of all villages around Koh Yor is over 90%.