A new hybrid equilibrium optimized SysFor based geospatial data mining for tropical storm-induced flash flood...

Area of Science:

• Hydrology
• Data Mining
• Natural Disaster Prediction

Background:

• Flash floods are dangerous hydrologic events with rapid onset, causing significant fatalities.
• Accurate spatial prediction of flash floods is challenging due to their complexity and precipitation dynamics.

Purpose of the Study:

• To propose a novel data mining approach, the hybrid equilibrium optimized SysFor (HE-SysFor) model, for enhanced spatial prediction of flash floods.
• To evaluate the HE-SysFor model's performance against other machine learning models in a tropical storm region.

Main Methods:

• Utilized a hybrid equilibrium optimized SysFor (HE-SysFor) model for flash flood spatial prediction.
• Applied the model to a case study in a tropical storm region in Northwest Vietnam, using 1866 flash-flooded locations and ten indicators.
• Compared HE-SysFor model performance with C4.5 decision tree, SVM-RBF, logistic regression, and DeepLNN models.

Main Results:

• The HE-SysFor model achieved high predictive performance: 93.8% total accuracy, 0.875 Kappa index, 0.939 F1-score, and 0.975 AUC.
• HE-SysFor outperformed C4.5, SVM-RBF, LReg, and DeepLNN models in both training and testing phases.
• Elevation, slope, and land cover were identified as the most significant indicators for flash flood prediction.

Conclusions:

• The HE-SysFor model offers a powerful alternative tool for effective flash flood monitoring in tropical areas.
• The findings support the development of robust policies for mitigating flash flood impacts.
• The study highlights the importance of key geographic indicators in flash flood prediction.