Feature-Optimized Multiclass Water Quality Detection in Kerala’s Medium Minor Rivers Using Siamese Networks

Water quality management is crucial to the sustainable development of river ecosystems. This system introduces a new classification system for the water quality levels of Kerala’s Medium Minor Rivers based on advanced machine learning techniques. The proposed methodology employs a hybrid data augmentation strategy known as Hybrid Augmentation for Water Quality Data, which integrates both numerical and categorical data augmentation to ensure a more diverse dataset. This new technique is the Graph-Contextual Embedding or GCE, how to extract meaningful features from the data by integrating graphbased feature extraction with context-aware similarity measures that enable it to capture complex relationships. For feature selection, the following bio-inspired optimization algorithms were used in order to find relevant features and remove redundancy: Spider Monkey Optimization (SMO), Vulture Search Algorithm (VSA), Whale Optimization Algorithm (WOA), Moth-Flame Optimization (MFO), and Elephant Herding Algorithm (EHA). A Siamese network is adopted at the classification step in order to learn discriminative feature embeddings comparing two different samples in the following way. The system classifies water quality into five categories: poor, medium, good, very good, and excellent. HAWQ + GCE + SMO + Siamese Network achieves an accuracy of $98.67 \%$, which indicates the effectiveness of the proposed method. This approach not only improves the classification of river water quality but also opens up possibilities for its application in broader environmental monitoring tasks. Future work will expand model performance and extend its use to other environmental domains.