A water ecological Internet of Things (IoT) system is comprised of numerous observing points equipped with diverse water quality IoT devices. It offers the feasibility for the exact prediction of water quality. However, conventional techniques encounter complexities regarding scalability, resource constraints and data privacy. Therefore, federated learning (FL) in IoT for water quality prediction improves data privacy, decreases communication costs, and allows decentralized and adaptive training. Here, the Fossa Greylag Goose Algorithm enabled a Deep Neural Xception Forward Fractional Network (FGGADNXFFNet) is proposed as a FL-based framework for predicting water quality within IoT environments. Here, the architecture consists of a central server and distributed IoT devices, where each device independently trains a local model using its own dataset. After local training, each IoT node sends its updates to the central server, where they are aggregated into a global model. Subsequently, the IoT nodes download the updated global model. Subsequently, training is updated in terms of downloaded local and global models. This procedure is repeated in each epoch until the optimal solution is achieved. The prediction of water quality is accomplished in a training model based on the following steps. Initially, the input data is normalized using decimal scaling normalization. Next, feature selection is done utilizing Jeffrey divergence. Lastly, water quality is predicted employing the Deep Neural Xception Forward Fractional Network (DNXFFNet). Next, training of DNXFFNet is done utilizing Fossa Greylag Goose Algorithm (FGGA). Furthermore, FGGA is devised by an integration of the Fossa Optimization Algorithm (FOA) with Greylag Goose Optimization (GGO). Additionally, FGGADNXFFNet has achieved 91. 683% of accuracy as well as False Positive Rate (FPR), loss, Normalized Mean Square Error (MSE), and normalized Root Mean Square Error (RMSE) of 0. 085, 0. 083, 0. 092 and 0. 304.
Londhe et al. (Fri,) studied this question.