A SMOTE PCA HDBSCAN approach for enhancing water quality classification in imbalanced datasets

Area of Science:

• Environmental Science
• Data Science
• Machine Learning

Background:

• Class imbalance is a major challenge in water quality classification, leading to biased models.
• Existing oversampling techniques often fail to adequately address noise and improve minority class performance.

Purpose of the Study:

• To introduce and evaluate SMOTE-PCA-HDBSCAN, a novel oversampling framework for imbalanced water quality datasets.
• To demonstrate the effectiveness of the proposed method in improving minority class sensitivity and overall classification accuracy.

Main Methods:

• The SMOTE-PCA-HDBSCAN framework combines Synthetic Minority Oversampling Technique (SMOTE), Principal Component Analysis (PCA), and Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN).
• SMOTE generates synthetic samples, PCA enhances data separability, and HDBSCAN removes noise from synthetic data.
• The cleaned synthetic data is merged with the original dataset for training.

Main Results:

• SMOTE-PCA-HDBSCAN significantly improved sensitivity for minority classes (Clean: 4.76% to 28.57%; Polluted: 38.09% to 61.90%).
• The method maintained high accuracy for the majority class.
• Comparative evaluations showed superior performance over other methods like SMOTE, SMOTE-DBSCAN, and SMOTE-ENN.

Conclusions:

• SMOTE-PCA-HDBSCAN is a robust and effective method for handling class imbalance in water quality classification.
• The framework offers a valuable tool for enhancing predictive models in environmental monitoring and other imbalanced data domains.