EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Predicting and Time‐Series Forecasting for Water Quality: Artificial Intelligence Approaches

Laila Amir

Applied Computational Intelligence and Soft Computing · 2026-01 · 2026 (1)

DOI: 10.1155/acis/6642357 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Applied Computational Intelligence and Soft Computing, online erschienen am 10.11.2026. Verfasst von Laila Amir. Die Arbeit nennt 50 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

While artificial intelligence (AI) is increasingly used for water quality assessment, high model performance does not necessarily imply independent predictive capability. A fundamental issue arises when Water Quality Index (WQI) classes are derived deterministically from the same physicochemical variables used as model inputs. In this setting, classification primarily measures the ability to reproduce predefined mathematical rules rather than to discover independent environmental patterns. Using a multidecadal dataset (1986–2020) from 84 monitoring stations in Hong Kong, this study evaluates machine learning (ML), deep learning (DL), and statistical approaches across three complementary tasks: deterministic WQI class reproduction, continuous WQI estimation, and long‐term forecasting. Classification robustness was further assessed under missing and noisy measurements. For forecasting, convolutional neural network (CNN), temporal convolutional network (TCN), and transfer learning–based bidirectional long short‐term memory (TL‐BiLSTM) models were optimized using a three‐stage hyperparameter framework and evaluated across 13 stations representing contrasting WQI regimes. For deterministic class reproduction, XGBoost and Random Forest (RF) achieved cross‐validation accuracies of 95.9% and 94.6%, respectively, compared with the theoretical WQI‐rule reference of 100%. Continuous WQI estimation also achieved high accuracy, with XGBoost obtaining root mean squared error (RMSE) = 2.056 and R 2 = 0.991 and RF obtaining RMSE = 2.490 and R 2 = 0.987. These results primarily indicate accurate approximation of the deterministic WQI formulation. Under degraded measurements, XGBoost and RF remained among the most robust models. Forecasting was substantially more challenging and strongly regime dependent: DL and ensemble approaches achieved the lowest RMSE at 8 of 13 stations, while statistical or naïve baselines performed best at 5 stations. In several cases, low RMSE was accompanied by weak or negative Pearson correlations, revealing a gap between numerical accuracy and temporal fidelity. Overall, the results show that WQI classification, continuous estimation, and temporal forecasting represent fundamentally different tasks. Model selection should, therefore, remain regime dependent and should always be evaluated against simple and deterministic reference baselines.

Zitieren

APA

Amir, L. (2026). Predicting and Time‐Series Forecasting for Water Quality: Artificial Intelligence Approaches. Applied Computational Intelligence and Soft Computing, 2026(1). https://doi.org/10.1155/acis/6642357

Vancouver

Amir L. Predicting and Time‐Series Forecasting for Water Quality: Artificial Intelligence Approaches. Applied Computational Intelligence and Soft Computing. 2026;2026(1). doi:10.1155/acis/6642357

BibTeX
@article{Amir2026,
  author = {Amir, Laila},
  title = {Predicting and Time‐Series Forecasting for Water Quality: Artificial Intelligence Approaches},
  journal = {Applied Computational Intelligence and Soft Computing},
  year = {2026},
  volume = {2026},
  number = {1},
  doi = {10.1155/acis/6642357}
}
Kostenlos starten →