A Hybrid VMD–Transformer–OCSVM Framework for Accurate Water Quality Anomaly Detection

Protecting aquatic ecosystems and public health requires accurate anomaly identification in water quality monitoring. This study suggests a brand-new hybrid anomaly detection framework called VMD–Transformer–OCSVM. which synergistically integrates Variational Mode Decomposition for signal denoising, a Transformer architecture for long-range temporal dependency modeling, and a One-Class Support Vector Machine for precise outlier classification. Experimental evaluation conducted on multivariate water quality datasets demonstrates that the proposed model significantly outperforms conventional approaches such as Isolation Forest and Autoencoders across key performance metrics. The hybrid model achieves superior precision (94.3%), recall (96.8%), F1-score (95.5%), and ROC-AUC (97.6%), while maintaining a notably low false positive rate of 3.5%. Although detection time is marginally higher, the enhanced interpretability through modal decomposition and attention mechanisms compensates for this trade-off. These findings confirm that the suggested framework for reliable, comprehensible, and highly accurate anomaly identification in intricate environmental monitoring systems is successful.