Metaheuristic algorithms applied in ANN salinity modelling

Salinity is a classic problem in planning the quality of freshwater resources management. Recent studies related to hybrid machine learning models have shown it's capability to simulate salinity dynamics. However, previous studies of metaheuristic algorithms have not dealt with comparing single- and hybrid-based algorithms in much detail. The present study aimed to develop univariate salinity by applying an artificial neural network model (ANN) integrated with (hybrid-based) coefficient-based particle swarm optimisation and chaotic gravitational search algorithm (CPSOCGSA). The methodology was developed and tested using electrical conductivity (EC) and total dissolved solids (TDS) data collected from the Euphrates River in Babylon Province, Iraq, from 2010 to 2019. The CPSOCGSA performance was evaluated by various single-based ones, including multi-verse optimiser (MVO), marine predator's optimisation algorithm (MPA), particle swarm optimiser (PSO), and the slim mould algorithm (SMA). The principal finding here confirms that hybrid-based outperformed four single-based algorithms based on different criteria. The outcomes for TDS were 0.004, 0.0248, and 0.98 for CPSOCGSA-ANN technique concern scatter index (SI), root-mean-squared error (RMSE), and correlation coefficient (R2), respectively. For EC, the results were 0.96 for R2, 0.0386 for RMSE, and 0.006 for SI. Due to its predictive accuracy, the proposed CPSOCGSA-ANN approach is suggested as a potential strategy for predicting monthly salinity data. Considering agriculture's vital role in Babylon Province's economy, this study may help inform future freshwater quality management decisions. •Artificial Neural Network technique's optimal hyperparameters and increase accuracy in predicting water salinity with less error.•CPSOCGSA-ANN (hybrid-based) performance was found to be better than other single-based models (i.e. SMA-ANN, PSO-ANN, and MVO-ANN).•A hybrid-based metaheuristic algorithm performs better than a single-based one.