Advanced IoT and Machine Learning Solutions for Sustainable Groundwater Management Using Edge-Based Residual Graph Attention Network Model

The sustainability of ecosystems and human resilience to changing and unpredictable climates are both impacted by groundwater, the world's biggest distributed freshwater storage system. Groundwater supplies can shift, nevertheless, due to abstraction, changes in land use, and climate change. Many traditional methodologies, such as targeted numerical approaches, have been suggested for groundwater modelling as potential answers to these problems. Preprocessing, feature selection, and training the model are the three stages that make up the suggested method. In pre-processing, the proposed approach addressed the null values, sparsity problem, and missing values to get the data ready for testing. In this stage, the proposed system apply a feature selection strategy that combines KM clustering with CFS ranking to deal with the high dimensionality of the input data. The suggested method relied on the Residual GAT for model training. Compared to state-of-the-art alternatives like GAT and LSTM, the proposed model typically achieves a better accuracy rate of 91.62%.