Construction of a VIKOR Composite Index-Based Probabilistic Prediction Model for Urban Water Blackening and Odor Recurrence Using Multi-Method Feature Selection
The construction of predictive models for the recurrence of blackening and odor in urban water bodies has become a critical foundation for refined management of urban water environments. Based on water quality monitoring data from 16 cities in the Yangtze River Basin from 2020 to 2023, this study systematically evaluated six typical comprehensive index calculation methods and proposed a probabilistic prediction model for water blackening and odor recurrence centered on the VIKOR composite index. Through sampling analysis and literature review, a power-law relationship between transparency (y) and turbidity (x) was established (y = 3.12x−0.66), leading to a critical turbidity threshold of 46.9 NTU for blackening and odor. Using ANOVA, recursive feature elimination, and random forest, five indicators—turbidity, dissolved oxygen (DO), total phosphorus (TP), permanganate index (CODMn), and ammonia nitrogen (NH3-N)—were selected as the model's indicator system, with importance ranking: turbidity > DO > TP > CODMn > NH3-N. The VIKOR composite index exhibited the most robust mapping relationship with blackening probability, achieving high accuracy (RMSE = 0.029, MAE = 0.020) and consistency (NSE = 0.918, R2 = 0.918), whereas models based on other indices yielded R2 values below 0.84. The model demonstrated good predictive performance across the Yangtze, Pearl, Haihe, and Yellow River basins. This model offers a universal decision-making tool for precise identification, early warning, and targeted management of water blackening and odor recurrence, with potential integration into urban water smart platforms.