Predictive modeling of groundwater quality near urban dump yards using N-BEATS and fuzzy inference systems

Groundwater quality near municipal dump yards is often degraded by leachate migration. This study analyzes and forecasts groundwater quality around the Kundrathur landfill in Chennai, India, using observations from 30 wells collected during the dry, monsoon, and post-monsoon seasons from 2020 to 2024. This research evaluates 13 physicochemical indicators, including pH, turbidity, DO, TDS, BOD, COD, TSS, chlorides, and sulfates, using an IoT/WSN buoy and laboratory validation. Further, this research also proposed a hybrid pipeline, a neural basis expansion analysis for interpretable time series forecasting (N-BEATS) neural network for initial temporal prediction with adaptive moment estimation optimizer (ADAM) for short-term time-series prediction, followed by a fuzzy inference system (FIS) to handle measurement uncertainty, and a Monte Carlo simulation for long-term probabilistic projections. N-BEATS (with Adam optimization) achieved 88% accuracy, MAE 0.0535, MSE 0.0105, Pearson correlation (r) 0.9502, RMSE 0.1024, and R² 0.8048, outperforming GRU (82% accuracy, MAE 0.0663, MSE 0.0110, r 0.9140, RMSE 0.1048) and LSTM (78% accuracy, MAE 0.0737, MSE 0.0124, r 0.9095, RMSE 0.1134). Short-term FIS outputs are provided for 387 days, while Monte Carlo projections explore multi-decadal behavior (e.g., layer-2 mean WQI ≈ 49.78 in the North and 48.89 in the East), consistent with a northeast-directed contamination trend visualized in QGIS. This work supports local groundwater protection and contributes to achieving SDG 3 (good health and well-being) and SDG 6 (clean water and sanitation).