Water Quality Forecasting Using a Hybrid Wavelet-ANFIS Model with Cross-Validation

Using the Adaptive Neuro-Fuzzy Inference System, this study assesses the precision of monthly water quality parameter training, validation, and prediction. The model analyzes historical data collected from continuous water quality monitoring stations along the Amudarya River, aiming to replicate secondary attributes (independent variables) from primary measurements (dependent variables). However, systematic and random errors introduce noise into the data, complicating the prediction process. To address this challenge, the study proposes an enhanced wavelet-based de-noising technique integrated with a data fusion framework for water quality prediction. The effectiveness of this approach was tested on key water quality parameters influenced by urbanization, including degree of conductivity of electricity, total amount of solids dissolved, and visibility. Results demonstrated that optimal prediction accuracy was achieved when cross-validation was set to two-sixths of the dataset. Additionally, the model outperformed the conventional Adaptive Neuro-Fuzzy Inference System model, significantly improving predictive accuracy. These findings highlight the proposed method as a promising alternative, presenting a computationally efficient algorithm with strong conceptual characteristics for de-noising and forecasting water quality. This method can support decision-makers in assessing water quality status and analyzing spatial and temporal variations.