Water quality describes the suitability of water in terms of its physical, chemical, and biological characteristics. To identify these characteristics adequately, one needs to comprehend the variation of the significant parameters affecting the water quality of a particular region. The study area for this work is the lower Mahanadi River basin, comprising of 13 stations located along the river course from which data from 20 different physicochemical parameters are collected for 2001 to 2023. Multivariate statistical methods are used to determine the seasonal variation of the water quality and to reduce the number of parameters. In particular, principal component analysis (PCA) provided desirable results that have a major impact on the water quality. Deep learning and machine learning have gained significant attention for analyzing time-series data. However, these methods often suffer from high complexity and significant forecasting errors, primarily due to non-linear datasets and hyperparameter settings. An innovative HDTO-DeepAR approach for predicting water quality indicators has been developed to address these challenges. HDTO-DeepAR outperformed the other methods. Improving surface water monitoring capabilities may result in accurate predictions, which can help policymakers develop a strategy to reduce water pollution. Traditional monitoring techniques are time-consuming and costly, making it difficult to meet the demands of real-time visualisation in current situations. To deal with this challenge, a novel approach (GHPSO-ATLSTM) has been developed to predict water quality indicators in surface water. The optimal features are selected using a genetic algorithm (GA), and the hyperparameters of LSTM are optimised with the hidden particle swarm optimisation (HPSO) technique followed by an attention (AT) layer to enhance the prediction accuracy.