Predicting and investigating water quality index by robust machine learning methods.

Area of Science:

• Environmental Science
• Data Science
• Urban Planning

Background:

• Accelerated urbanization intensifies environmental degradation and public health risks in urban areas.
• Effective waste management and water quality monitoring are critical for sustainable urban development.
• Existing methods often predict continuous water quality values, lacking nuanced classification.

Purpose of the Study:

• To predict the Water Quality Index (WQI) in urban environments using advanced machine learning algorithms.
• To classify WQI into discrete labels representing water quality from "good" to "poor".
• To compare the performance of Long Short-Term Memory (LSTM), Random Forest (RF), Decision Tree (DT), and Support Vector Machine (SVM) for WQI prediction.

Main Methods:

• Integration of multiple machine learning algorithms (LSTM, RF, DT, SVM) into a unified framework.
• Classification of WQI into 9 distinct labels (1-9) instead of continuous value prediction.
• Utilizing training, testing, and validation datasets to evaluate predictive accuracy and precision.

Main Results:

• Long Short-Term Memory (LSTM) demonstrated superior predictive accuracy and precision compared to RF, DT, and SVM.
• LSTM achieved low RMSE values (e.g., 0.0611 on training data) and R² values consistently above 0.9964.
• The model effectively captured complex temporal dependencies in water quality data.

Conclusions:

• LSTM is a robust and reliable tool for urban water quality prediction, surpassing other tested algorithms.
• The classification approach provides actionable insights for identifying pollution factors and optimizing waste management.
• This research offers a scalable and practical solution for improving urban environmental management and public health outcomes, considering climate change impacts.