Novel framework of water quality prediction using improved optimization-based multi-cascaded deep serial network

Water quality prediction is crucial for the environmental system to optimally enhance the productivity of the agricultural field and protect people from various health issues. However, the inherent complexity of aquatic environments introduces significant noise into time-series data, reducing the reliability of traditional forecasts. Here, an effective water quality prediction framework is introduced in this paper with the aid of deep learning. At first, the data are collected from standard datasets. Further, the collected data are forwarded to the Adaptive Multi-cascaded Deep Serial Network (AMDSNet) for water quality prediction. The introduced AMDSNet is incorporated with the autoencoder, convolutional neural network (CNN) and Bayesian learning (BL). In addition, the Modified Running City Game Optimizer (MRCGO) is designed in the optimization process to quickly tune the parameters of AMDSNet. Finally, the overall performance of the designed framework is validated with various performance measures. Further, the outcomes are compared with various conventional methods to demonstrate their effectiveness. In the batch size validation, the proposed approach has attained 96% and 97% values in Dataset 1 and Dataset 2 at the 100th batch size. Additionally, the developed approach achieved a lower MAE value of 0.07 at a 500th batch size.