Machine Learning Approach for Water Pollution Detection
Loading...
Date
2026-06-07
Journal Title
Journal ISSN
Volume Title
Publisher
Larbi Tebessi University-Tebessa
Abstract
Water pollution remains a critical environmental and public health challenge, while
traditional monitoring approaches based on manual sampling and laboratory analysis
are often costly, limited in coverage, and unsuitable for real-time detection. This thesis
proposes a machine learning framework for automated water pollution detection using
ensemble learning, feature engineering, and adaptive class imbalance handling. Three
benchmark datasets of increasing complexity are investigated: binary water safety classi-
fication, multi-class pollution severity prediction, and Industrial IoT time-series anomaly
detection.
The proposed experimental pipeline evaluates models under both raw and resampling-
augmented training conditions, combining imbalance mitigation techniques with Bayesian
hyperparameter optimization. Experimental results show that ensemble models, particu-
larly gradient-boosting methods, consistently achieve strong predictive performance across
all datasets and outperform the benchmark deep learning approach used for comparison.
An important finding of this study is that synthetic oversampling techniques may
reduce performance when applied to engineered feature spaces or models with built-in cost-
sensitive learning capabilities. Overall, the results demonstrate that optimized ensemble
methods combined with principled feature engineering provide a robust, interpretable,
and scalable solution for intelligent water quality monitoring systems.
Description
Keywords
Water pollution detection, machine learning, ensemble methods, class im- balance, anomaly detection, water quality classification, feature engineering, SMOTE, gradient boosting, IoT sensor data