Machine Learning–Based Assessment of Environmental and Public Health Risks in Sewerage Sludge Management Systems

Abstract

This quantitative study examined machine learning–based assessment of environmental and public health risks in sewerage sludge management systems using sludge quality measurements, contaminant concentration data, treatment performance indicators, and public health risk variables. The study adopted a quantitative predictive design and analyzed a final dataset of 250 observations collected from selected sludge management facilities after 18 incomplete records and 7 duplicate observations were removed during data screening. Descriptive analysis showed substantial variation across the measured variables, with mean heavy metal concentration recorded at 178.42 mg/kg, mean organic pollutant concentration at 91.35 mg/kg, mean moisture content at 68.52%, and mean treatment efficiency at 83.67%. The average environmental risk score was 61.24, while the mean public health risk index was 58.13, indicating moderate overall risk across the sampled facilities. Comparative analysis across sludge treatment categories showed that primary sludge produced the highest mean environmental risk score at 72.45, whereas treated biosolids produced the lowest mean score at 47.63. Regression analysis demonstrated that treatment efficiency, heavy metal concentration, pathogen indicator score, and organic pollutant concentration were significant predictors of environmental and public health risk outcomes. The model explained 74.2% of the variance in risk scores, confirming strong predictive power. Machine learning analysis indicated that the Random Forest model achieved the highest predictive performance, with 93.8% testing accuracy, 92.6% precision, 94.2% recall, an F1-score of 93.4%, and an AUC of 0.961. Artificial Neural Network and Support Vector Machine models also demonstrated strong performance, with testing accuracies of 91.7% and 89.9%, respectively. Feature importance analysis identified treatment efficiency as the strongest predictor, contributing 28.4% to model performance, followed by heavy metal concentration at 24.7% and pathogen indicator score at 18.3%. Overall, the findings confirmed that machine learning models provided reliable quantitative tools for classifying environmental and public health risks in sewerage sludge management systems.