Identification of the best model to predict optical properties of water

Abstract

This research studies prediction models for optical water properties, which are very important to save time consumption to easy control of irrigation systems to make it smart irrigation systems and changing manual system with automatic systems, moreover the suitable time to implement maintenance. One of the most important reasons to use a regression model in this study is to have the ability to use a simple laser device to measure one optical parameter or physical parameters to predict the other parameters by regression models, which will save more money used by an expensive laser device to measure the three optical properties (reflection, transmission, and absorption). Data collected through laboratory experiments were then put through various machine learning regression models. The applied models are random forest (RF), XGBoost (XGB), support vector machine (SVR), and multiple linear regression (MLR). Four scenarios implemented these models; the first scenario included [pressure difference (Bar), wavelength (nm), total suspended solids (TSS)] for samples of Water taken from the filtration system, the second scenario included (Bar, nm, TSS, and reflection), the third scenario included (Bar, nm, TSS and transmission). The fourth scenario is (Bar, nm, TSS, transmission, and reflection). The findings of this research were SVR in scenario 3 (SVR3) had a high correlation coefficient (R2 = 0.95) and minimum error, followed by MLR in scenario 3 (MLR3) and scenario 2 (MLR2), which had the same correlation coefficient (R2 = 0.95), but errors were higher than SVR3. The lowest root means square error (RMSE) was in SVR in scenario 1 (SVR1) by 0.180, followed by RF in scenario 1 (RF1) by 0.181, and the highest was in RF in scenario 2 (RF 2) by 15.2. Based on these performance criteria, SVR3 is the best model to predict optical properties in this study.