Feature selection methods for characterizing and classifying adaptive Sustainable Flood Retention Basins.

Area of Science:

• Environmental Science
• Water Resource Management
• Data Science

Background:

• The EU Flood Directive mandates flood risk mapping, necessitating efficient assessment tools.
• Sustainable Flood Retention Basins (SFRBs) are crucial for flood defense and pollution control.
• Existing SFRB assessment methods involve over 40 variables, which are time-consuming and often correlated.

Purpose of the Study:

• To explore correlations among SFRB characterization variables.
• To identify the most important variables for effective SFRB classification.
• To develop a rapid and accurate SFRB assessment framework.

Main Methods:

• Applied three feature selection techniques (Information Gain, Mutual Information, Relief) to an SFRB dataset.
• Utilized four benchmark classifiers (SVM, KNN, C4.5, Naïve Bayes) to verify classification effectiveness.
• Identified the optimal subset of variables for SFRB classification.

Main Results:

• Identified a significantly reduced set of nine important variables sufficient for accurate SFRB classification.
• Demonstrated the proposed approach provides a simple, rapid, and effective framework for variable selection.
• Verified the performance of the nine selected variables across six typical SFRB cases.

Conclusions:

• A simplified SFRB assessment using only nine key variables is feasible and accurate.
• The findings offer a rapid scientific tool for practical SFRB assessment and flood risk management.
• The developed methodology has broad applicability in other assessment domains.