Research Article Open Access

Spatio-Temporal Anomaly Detection in Groundwater Electrical Conductivity Using a Hybrid Framework of Isolation Forest and Autoencoder

Eunji Lee1, Seunghyun Lim2, Seojun Lee3 and Abhijit Debnath4
  • 1 Branksome Hall Asia, 234 Global Rdu-Ro, Daejeong-Eup, Seogwipo-Si, Jeju-do, Korea
  • 2 Seoul International School, 15 Seongnam-Daero, Gyeonggi-do, Korea
  • 3 Korean International School- Pangyo Campus, Gyeonggi-do, Korea
  • 4 Department of Computer Science and Engineering, Institute of Engineering and Management, Newtown, University of Engineering and Management, Action Area-III, Newtown Road, Kolkata, India

Abstract

Monitoring groundwater quality is vital for environmental safety and resource sustainability. This study combines Isolation Forest and Autoencoder models to detect anomalies in Electrical Conductivity (EC) and temperature, using monthly data collected in South Korea between 2006 and 2023. Linear regression and the Mann-Kendall test reveal a weak, episodic downward EC trend. Seasonal decomposition indicates annual cyclicality, while residual analysis uncovers localized anomalies. K-means clustering differentiates normal and contaminated groundwater patterns. The results highlight the effectiveness of integrating statistical and machine learning approaches for interpretable, data-driven groundwater quality monitoring in data-scarce environments.

Journal of Computer Science
Volume 22 No. 1, 2026, 273-383

DOI: https://doi.org/10.3844/jcssp.2026.273.383

Submitted On: 30 April 2025 Published On: 14 February 2026

How to Cite: Lee, E., Lim, S., Lee, S. & Debnath, A. (2026). Spatio-Temporal Anomaly Detection in Groundwater Electrical Conductivity Using a Hybrid Framework of Isolation Forest and Autoencoder. Journal of Computer Science, 22(1), 273-383. https://doi.org/10.3844/jcssp.2026.273.383

  • 48 Views
  • 5 Downloads
  • 0 Citations

Download

Keywords

  • Groundwater Quality Monitoring
  • Anomaly Detection
  • Unsupervised Learning
  • Seasonal Decomposition
  • Hybrid Machine Learning Models