Work

A Class-Adaptive Hybrid Mean Aggregation Framework for River Water-Quality Classification: Resolving Eclipsing and Ambiguity in the Malaysian DOE-WQI

Year: 2026
Type: preprint
Abstract: Conventional aggregation operators in Water Quality Index (WQI) systems such as the weighted arithmetic mean used by the Malaysian Department of Environment (DOE-WQI) is suffering from problem in eclipsing and ambiguity. To resolve this problem, this study proposes a class-adaptive Hybrid Mean Function that applies band specific hybrid aggregation operators across DOE water-quality categories (Clean, Slightly Polluted, Polluted). Within a generalized power mean equation with eight model variants (A–H) representing a \(\:{2}^{3}\) factorial combination of min or max operators were evaluated under both equal and unequal weighting schemes. The framework was validated using 630 observations across six parameters (DO, BOD, COD, SS, AN, pH) from Malaysian rivers from year 2017 to 2021and benchmarked against DOE-WQI alongside three PCA-derived variants. Robustness was evaluated via non-parametric classification metrics, bootstrap resampling (B = 1000), and PCA sensitivity analysis. Model F (min, max, max under unequal weighting) emerged as uniquely robust, achieving a mean class-similarity of 80.8%, a macro-F1 score of 0.785, and an overall accuracy of 0.860 is outperforming the next-best model in the Slightly Polluted class by 5.9 percentage points. By involving min in the clean class and max in polluted class, Model F simultaneously reduces eclipsing and ambiguity problem. This shows a mathematically transparent, highly accurate alternative for river monitoring and sub-basin prioritization in Malaysia.
Language: English
Open Access status: green
OpenAlex ID: W7235703628
Cites: 17
Cited by:
Related to: 10