Predicting Colour Index of Industrial Effluent Using Correlation-Based Forward Selection Regression and Principal Component Regression

Water quality monitoring is crucial for ensuring compliance with environmental regulations and supporting sustainable industrial practices. The current methods for assessing the colour index depend heavily on labour-intensive laboratory techniques, which can be time-consuming and costly. This study investigated the use of machine learning approaches to predict the American Dye Manufacturers Institute (ADMI) colour index (colourADMI) of textile industrial effluent from manufacturing processes. Two machine learning algorithms, i.e. Correlation-Based Forward Selection Regression (C-FSR) and Principal Component Regression (PCR), with varied normalization techniques, were implemented to predict the colourADMI based on other wastewater quality parameters, i.e., chemical oxygen demand (COD), pH, chromium hexavalent (Cr(VI)), copper (Cu(II)), suspended solids (SS), total dissolved solids (TDS), and raw colour (colourRaw). First, three different normalization techniques were applied to investigate the effects of normalization. Then, Multiple Linear Regression (MLR) was applied with a forward selection procedure, which began with the most correlated independent variable and progressively adding others to perform C-FSR. This process was used to identify the optimal C-FSR. After that, MLR coupled with Principal Component Analysis (PCA) was performed to establish Principal Component Regression (PCR) models to eliminate the effect of collinearity among independent variables. The results indicate that normalization techniques had minimal impact on C-FSR performance, while C-FSR that used five independent variables, demonstrated better predictive accuracy than PCR that used four principal components (PCs). Hence, C-FSR with five variables was considered the optimal method due to its simplicity and high predictive accuracy. This highlights the importance of selecting the optimal input variables for more accurate colourADMI predictions.