Machine Learning Based Alum Dosing Optimization for Adaptive Water Quality Management in Treatment Plant

MD. Minaruzzaman Shovon , H A Hossain Tamjid , Md. Tamim

Ensuring safe and cost-effective water purification remains
a critical challenge, particularly for large natural water
bodies like the Halda River, where water quality parameters
fluctuate significantly. Traditional methods for determining
alum dosages often rely on manual experiments that fail to
adapt to real-time variations, leading to inefficiencies and
chemical overuse. This study presents a data-driven solution
to optimize alum dosage prediction using advanced machine
learning techniques. A comprehensive dataset comprising raw
and treated water quality parameters including turbidity, pH,
alkalinity, and chloride from the Halda River was utilized.
Multiple models were trained including Random Forest,
Gradient Boosting, Support Vector Machine, Logistic
Regression, and K-Nearest Neighbors. The KNN model achieved
the highest performance with a test accuracy of 94.87% and
an ROC AUC of 0.957. Feature importance analysis revealed
that turbidity, chloride concentration, and pH-related
interactions are the most influential predictors of alum
demand. The developed model offers a practical framework for
real-time automated alum dosage recommendations, enabling
water treatment facilities to reduce operational costs,
minimize chemical waste, and improve treatment consistency.