Statistical Techniques for Sewage Wastewater Treatment Prediction and Optimization via Cucumis Melo Coagulant

The wastewater treatment process involves multiple stages; physical, chemical, and biological designed to remove contaminants such as organic matter, nutrients, pathogens, and heavy metals. Integrating artificial intelligence, statistical modeling, and optimization algorithms enhances prediction and control mechanisms, leading to more sustainable and cost-effective wastewater management solutions. Previous studies optimized multiple responses (turbidity, total suspended solids (TSS), biochemical oxygen demand (BOD), and chemical oxygen demand (COD)) using two operational factors (pH and coagulant dosage) at low and high levels (-1, +1) within a quadratic regression model (QM). However, these efforts were limited by factor selection, narrow variable bandwidths, and the chosen regression model, resulting in insufficient data fitting. This paper introduces an additional critical factor and expands the factor levels to five (-α, -1, 0, 1, +α) to enhance data fitting and optimize responses based on process requirements. An adaptive regression model is applied to improve goodness-of-fit statistics, increase dispersion from the zero-residual line, and enhance optimization outcomes. Comparative analysis, including 3D surface plots, demonstrates that the local linear regression model with variable bandwidths outperformed the QM model in design efficiency, delivering superior goodness-of-fit, reduced residual errors, and better optimization results, with an overall desirability of 100% compared to 95.50% for QM using Cucumis Melo as a natural coagulant. Keywords: Local linear regression model, Cucumis Melo, Turbidity, Total Suspended Solids, Biochemical Oxygen Demand, Chemical oxygen demand