Abstract: Random Forest (RF) classifies and predicts data and, subsequently, determines data patterns, particularly water quality. This pattern was identified, using data from secondary sources analyzed with R software, by the RF technique and wastewater parameters (i.e., date, turbidity, total suspended solids, total coliform, pH, electrical conductivity, chemical oxygen demand, and biochemical oxygen demand). Statistical methods were then used to analyze the data, which was further refined using the machine learning method with hyperparameter optimization that constructs multiple independent trees to increase prediction precision. To ensure that the prediction error converges, a sufficiently large number of trees is needed. The hyperparameter grid models were optimized, and the tuned model with an RMSE of 34.86852 (3.2% improvement over out-of-bag performance RMSE of 36.02448) ranked first. The formulation of the RF model is capable of detecting and estimating the nonlinear inter-relationships of the components (i.e., water clarity, pH acidity, and total suspended solids) that ought to be analyzed in determining the quality of wastewater in treatment plants.

Keywords: Random Forest, Machine learning, Wastewater quality, Hyperparameters, Treatment plant

Pages: 103-113