Artificial Intelligence and Machine Learning in Industrial Wastewater Treatment: Bibliometric Trends, Application Areas, and Critical Gaps (2016–2025)

The growing complexity and variability of industrial wastewater (IWW) require treatment strategies that can support reliable prediction, monitoring, optimization, and operational decision-making. Artificial intelligence (AI) and machine learning (ML) have increasingly been applied to these functions; however, the literature remains fragmented across pollutants, treatment processes, and modeling approaches. This study presents a critical bibliometric review of 151 Scopus-indexed publications on AI/ML applications in IWW treatment published between 2016 and 2025. Publication output increased substantially, reaching 35 publications in 2024 and 67 in 2025, compared with 1–4 publications annually during 2016–2019. China was the leading contributor with 54 publications, followed by India with 26, while Saudi Arabia and the United States each contributed 15. The main research directions included prediction and soft sensing, process optimization and control, monitoring and digitalization, adsorption and material-based treatment, membrane processes, and advanced oxidation and degradation. Artificial neural networks, support vector machines, random forests, and hybrid models were among the most frequently applied techniques. Study-level assessment of 38 technically assessable original AI/ML applications showed that 34 studies (89.5%) relied on offline or retrospective model evaluation, whereas only four (10.5%) demonstrated online or real-time AI/ML evaluation or operation. Conventional internal validation predominated among studies for which validation design could be established, while independent temporal, cross-domain, cross-scale, or field-based validation remained comparatively uncommon. Laboratory- and bench-scale applications were common, although several studies used full-scale industrial data. Major barriers to practical deployment include limited transferability across wastewater matrices, insufficient field-scale validation, inconsistent sensor and data quality, limited interpretability, and weak integration with treatment-plant control systems. Chemical oxygen demand soft sensing, pH and chemical-dosing support, and real-time monitoring and anomaly detection represent promising pathways for near-term implementation, while closed-loop control requires stronger validation of reliability, safety, and long-term performance. Overall, the field is transitioning from isolated, model-centered research toward system-level deployment, where transferability, validation, interpretability, and control integration are as important.