Optimization of UV-C irradiation time for biologically treated saline wastewater using artificial neural networks and a hybrid multi-criteria decision-making framework

Biologically treated saline industrial wastewater may still contain residual organic matter, nutrients, petroleum hydrocarbons, and dissolved constituents that limit the quality and potential reuse of the treated effluent. Ultraviolet-C (UV-C) irradiation is a chemical-free polishing technology with potential for further improving effluent quality; however, its treatment performance depends strongly on irradiation time, and excessive exposure may increase energy consumption without providing proportional improvements. Moreover, optimization based on a single water-quality parameter cannot adequately represent the simultaneous changes and trade-offs among multiple effluent-quality indicators. Existing approaches commonly evaluate UV treatment using individual quality parameters or conventional statistical analyses, while artificial neural networks provide prediction capability but have limited interpretability for selecting an operating condition. Conversely, multi-criteria decision-making (MCDM) approaches can support transparent optimization but are generally restricted to experimentally investigated conditions. Therefore, an integrated framework that combines experimental UV-C treatment, predictive modeling, and interpretable multi-criteria optimization is required for systematic determination of the irradiation time. This study aimed to experimentally investigate the effect of UV-C irradiation time on biologically treated saline industrial wastewater and to develop an integrated ANN–MCDM framework for predicting wastewater-quality responses, identifying influential quality parameters, and determining the optimum irradiation condition. A laboratory-scale reactor equipped with a 30 W low-pressure mercury lamp emitting at 254 nm was employed at irradiation times of 40–320 s. Prior to irradiation, 4.8 g of dry ice was used for pH conditioning. Nine wastewater-quality parameters, namely COD, BOD, phosphate, nitrate, dissolved organic carbon (DOC), total petroleum hydrocarbons (TPH), turbidity, total dissolved solids (TDS), and pH, were evaluated. A 64–32–16–9 multilayer perceptron ANN was developed for simultaneous prediction of the normalized quality parameters. In parallel, Min–Max normalization, K-Means clustering, principal component analysis (PCA), a Composite Wastewater Quality Index (CWQI), TOPSIS, and sensitivity analysis were integrated to classify treatment regimes, determine parameter weights, evaluate overall treatment quality, rank irradiation times, and identify the most influential quality indicators. Wastewater quality generally improved with increasing irradiation time, with the most pronounced overall improvement occurring between 240 and 280 s, followed by a saturation tendency at longer exposure. The integrated analysis identified 280 s as the optimum irradiation time , yielding a maximum CWQI of 0.977 and a TOPSIS score of 0.966 . At this condition, COD and BOD reductions reached approximately 41.1% and 17.5% , respectively, while improvements were also observed in DOC, TPH, phosphate, nitrate, turbidity, and TDS. At 320 s, the CWQI and TOPSIS score decreased to 0.925 and 0.894 , respectively, indicating that prolonged irradiation did not provide a corresponding improvement in overall treatment performance. PCA identified phosphate as the most influential parameter, with a normalized weight of 0.168 , followed by COD ( 0.149 ), DOC ( 0.139 ), and BOD ( 0.125 ). The sensitivity analysis produced the same parameter-importance sequence. The ANN achieved an average test RMSE of approximately 0.12 on the normalized scale and provided predictions for irradiation times not explicitly included in the experimental dataset. The ANN and hybrid MCDM approaches provide complementary capabilities: the ANN enables simultaneous nonlinear prediction at untested irradiation times, whereas the hybrid framework provides greater mathematical transparency through clustering, objective weighting, composite quality assessment, ranking, and sensitivity analysis. The agreement between CWQI and TOPSIS in identifying 280 s demonstrates consistency between two independent multi-criteria evaluation approaches. From a practical perspective, the results indicate that extending irradiation beyond the optimum does not necessarily improve overall effluent quality and may increase irradiation time and associated energy demand without proportional treatment benefits. The proposed integration of UV-C polishing, ANN-based prediction, and an interpretable hybrid MCDM framework provides a systematic approach for evaluating and optimizing biologically treated saline industrial wastewater. The methodology enables simultaneous consideration of multiple water-quality parameters while retaining predictive capability for untested operating conditions. The identified optimum of 280 s provides a data-driven operating condition for the investigated laboratory system and establishes a basis for further studies involving repeated experiments, uncertainty analysis, explicit UV fluence measurements, microbial inactivation, and scale-up toward practical wastewater reuse applications.