<p>To achieve sustainability in the water treatment processes through electrocoagulation, optimal operating parameters are required for its proper functioning and to fulfil environmental goals. Further, optimal values for electrocoagulation depend on accurate physical and numerical models. These models simulate the complexities of the treatment process. This study explores the capabilities of the deep learning modelling tool artificial neural network (ANN) for multi-objective modelling. ANN models the removal of arsenic and fluoride from the water with respect to current density, pH, time, and initial concentrations of arsenic (As) and fluoride (F). This study indicates that ANN models have higher accuracy than isotherm models for representing the electrocoagulation process as reported higher coefficient of determination values of ANN (As:0.9995, F:0.9990) over isotherm models (Langmuir, Freundlich and Sips Models). To achieve the sustainability and efficiency of the electrocoagulation process, the best model, in this case, the ANN model, used to find optimal values whereby effective remediation of the arsenic and fluoride from the water can be achieved. For the same, multiple metaheuristic optimisation tools are explored on the objective function created from the ANN model. This study explored genetic algorithm (GA), particle swarm optimisation (PSO), and ant colony optimisation (ACO) meta-heuristic tools and compared these tools on economic terms. Upon validation of these optimal conditions, it was reported that PSO (2.62 mAcm⁻², pH: 6.5, 45&#xa0;min, F: 10mgL⁻¹ and As: 450 µgL⁻¹) showed more correlated and economical operating conditions (1.584USDm⁻³) to achieve the 96% fluoride removal and 97% arsenic removal.</p> Graphical Abstract <p></p>