Estimating chemical oxygen demand in municipal landfill leachate using multilayer perceptron artificial neural networks based on seasonal monitoring data

Landfill leachate is a hazardous by-product of municipal solid waste (MSW) and chemical oxygen demand (COD) is a key indicator for evaluating its pollution strength and treatment needs. This study aimed to predict leachate COD concentrations from the Nigde Municipal Solid Waste Landfill Site using a Multilayer Perceptron Artificial Neural Network (MLPANN). 52 weekly leachate samples were collected and analyzed for physicochemical indicators, including pH, the ambient temperature, total solids (TS), oil and grease (OG), electrical conductivity (EC), arsenic (As), cobalt (Co), and cadmium (Cd). The principal component analysis (PCA) identified three significant components that explained 77.84% of the total variance. pH, temperature, Cd, Co, and OG had significant loading scores among other parameters. Four models were tested with different input selections. Before running the models, the data set (n = 52) was split with season strata to 70% as the training set and 30% as the testing set. After cross-validation, the best model was selected based on the lowest error metrics. The best-performing model, which incorporated variables selected via PCA performed the best during cross-validation and generalization. Its final architecture (5-21-1) was used for the testing set and achieved a correlation coefficient of 0.864. This study also represents the first application of the brulee engine within the tidymodels framework for leachate COD prediction, offering a reproducible modeling approach for environmental monitoring studies.