Enhancing slope stability prediction through transfer learning based on two-stage TrAdaBoost

Accurate prediction of slope stability is crucial for engineering safety and disaster prevention. Traditional physical and numerical approaches often face limitations when dealing with complex conditions and require substantial resources. To address this, we propose a two-stage TrAdaBoost-based model that integrates transfer learning with machine learning algorithms (DT, RF, SVM, GBT, XGBoost) for slope stability prediction. The key advantage lies in expanding sample size by leveraging source-domain data, thus improving prediction accuracy for the target domain and overcoming the limitations of small datasets. Unlike conventional machine learning models that rely on the assumption of independent and identically distributed (i.i.d.) data, our method relaxes these assumptions through dynamic weight adjustment and domain adaptation. Evaluated using MSE, MAE, and MAPE, the proposed model demonstrates consistently superior predictive performance, particularly with SVM (MSE: 0.0185, MAE: 0.1059, MAPE: 8.67%) and XGBoost (MSE: 0.0197, MAE: 0.1090, MAPE: 8.80%), outperforming traditional DT, SVM, and DNN methods. These findings demonstrate the feasibility of the proposed approach, which not only enhances slope stability prediction under limited data but also offers potential applications in landslide forecasting and disaster risk reduction.