Machine Learning Models for Predicting Water Quality of Treated Fruit and Vegetable Wastewater

Abstract

Wash-waters and wastewaters from the fruit and vegetable processing industry are characterized in terms of solids and organic content that requires treatment to meet regulatory standards for purpose-of-use. In the following, the efficacy of 13 different water remediation methods (coagulation, filtration, bioreactors, and ultraviolet-based methods) to treat fourteen types of wastewater derived from fruit and vegetable processing (fruit, root vegetables, leafy greens) were examined. Each treatment was assessed in terms of reducing suspended solids, total phosphorus, nitrogen, biochemical and chemical oxygen demand. From the data generated, it was possible to develop predictive modeling for each of the water treatments tested. Models to predict post-treatment water quality were studied and developed using multiple linear regression (coefficient of determination (R2) of 30 to 83%), which were improved by the generalized structure of group method of data handling models (R2 of 73–99%). The selection of multiple linear regression and the generalized structure of group method of data handling models was due to the ability of the models to produce robust equations for ease of use and practicality. The large variability and complex nature of wastewater quality parameters were challenging to represent in linear models; however, they were better suited for group method of data handling technique as shown in the study. The model provides an important tool to end users in selecting the appropriate treatment based on the original wastewater characteristics and required standards for the treated water.

Keywords

• water quality
• multiple linear regression (MLR)
• wastewater
• machine learning
• wash-water
• food processing