Flash flood susceptibility modeling using an optimized fuzzy rule based feature selection technique and tree based...

Area of Science:

• Environmental Science
• Geospatial Analysis
• Data Science

Background:

• Flash floods pose significant risks, necessitating accurate susceptibility modeling.
• Traditional methods often lack the precision required for effective flood risk management.
• Feature selection and ensemble methods offer potential for improved predictive performance.

Purpose of the Study:

• To develop and evaluate a novel methodological approach for flash flood susceptibility modeling.
• To integrate a feature selection method (FURIA-GA) with tree-based ensemble algorithms (LogitBoost, Bagging, AdaBoost).
• To assess the predictive performance of the developed models in Vietnam's Lao Cai Province.

Main Methods:

• A feature selection method (FSM) combining FURIA (fuzzy rule-based algorithm) and Genetic Algorithms (GA) was employed.
• The FURIA-GA FSM was integrated with LogitBoost, Bagging, and AdaBoost ensemble algorithms.
• Model performance was evaluated using classification accuracy, sensitivity, specificity, and Area Under the Curve (AUC).

Main Results:

• The FURIA-GA FSM yielded more accurate predictive results than conventional methods.
• FURIA-GA-Bagging achieved the highest predictive classification accuracy (93.37%), sensitivity (96.94%), and specificity (89.80%).
• FURIA-GA-AdaBoost recorded the highest prediction AUC value (0.9740), indicating superior predictive ability.

Conclusions:

• The proposed FURIA-GA based ensemble models offer a novel and effective approach for flash flood susceptibility mapping.
• The choice of statistical metrics can influence the determination of the best predictive model, highlighting site-specific factors.
• These advanced models provide valuable tools for environmental risk assessment and disaster management planning.