Multi-source data-driven prediction of cold-region slope failure using an SSA-PNN optimized stepwise reduction...

Area of Science:

• Geotechnical Engineering
• Machine Learning Applications
• Cold Region Engineering

Background:

• Cold-region slopes face instability due to freeze-thaw cycles degrading geomaterials.
• Conventional methods struggle to capture progressive failure mechanisms in these slopes.
• Accurate prediction of slope degradation is crucial for geohazard risk assessment.

Purpose of the Study:

• To develop an integrated framework for analyzing slope instability under freeze-thaw conditions.
• To improve the characterization of spatiotemporal slope degradation mechanisms.
• To enhance the reliability of intelligent monitoring and early warning systems for cold-region geohazards.

Main Methods:

• Integrated Stepwise Reduction Method (SRM) with a Sparrow Search Algorithm-optimized Probabilistic Neural Network (SSA-PNN).
• Utilized a multi-source dataset including field monitoring, numerical simulations, and laboratory tests.
• Evaluated model performance using classification and regression tasks with safety factor (FS) as the label.

Main Results:

• The SSA-PNN model achieved high accuracy (87.5% classification, R²=0.871 regression).
• Outperformed benchmark models like XGBoost, SVM, and Logistic Regression.
• Significantly reduced misclassification rates in critical stability intervals compared to conventional PNN.

Conclusions:

• The SRM-SSA-PNN framework effectively models freeze-thaw-induced slope degradation.
• The framework enhances the interpretability of slope instability mechanisms.
• Provides a reliable basis for risk assessment and early warning of geohazards in cold regions.