Data mining algorithms for the prediction of inorganic nitrogen in water: Mechanisms, techniques, and applications

A summary of the data mining algorithms and their applications covered in this review over the past fifteen years. TN: total nitrogen. • A comprehensive review data mining algorithms advances for inorganic nitrogen prediction in water. • Essential data preparation methods tailored for inorganic nitrogen prediction were synthesized. • Comparative analysis of different model families revealed deep learning as the most promising approach. • Key challenges and future directions of data mining for inorganic nitrogen prediction were critically analyzed. Accurate prediction of inorganic nitrogen concentrations is critical for assessing water quality and supporting sustainable aquatic ecosystem management. Data mining algorithms have demonstrated significant potential in inorganic nitrogen prediction. This review provides a systematic examination of data mining algorithms applied to inorganic nitrogen prediction for aquatic environments, focusing on the complete modeling pipeline from data preparation to predictive algorithm deployment. Essential data preparation techniques involving data acquisition, data cleaning, dimensionality reduction, and dataset partitioning are described in detail. Then, the mechanisms, techniques, and applications are critically evaluated according to the following five categories of data mining algorithms: machine learning models, deep learning models, tree-based models, fuzzy neural networks, and hybrid models. FNN and deep learning models yield top-tier accuracy (mean R 2 > 0.90). In addition, hybrid models integrating LSTM have emerged as the most promising and prevalent framework, driving advancements in prediction precision. Despite significant progress, the field continues to face challenges such as insufficient data quality, limited model generalizability, and a lack of interpretability. Integrating multimodal data, developing lightweight real-time prediction systems, combining explainable artificial intelligence with physics-informed constraints, and strengthening uncertainty quantification will strongly advance inorganic nitrogen prediction models toward more accurate, reliable, and practical decision-support tools. This review provides a systematic reference for researchers and practitioners.