<p>A three-dimensional monitoring data-driven updating parameter back-analysis and slope stability analysis method is proposed to quantitatively and efficiently assess the effectiveness of anti-slide piles for colluvial slope stabilization. By integrating the multi-output Gradient Boosting Decision Tree (GBDT), a meta-model is constructed to characterize the relationship between geotechnical parameters and monitored displacements. This meta-model enables deterministic and probabilistic back-analyses based on multiple monitoring points. Another meta-model is developed to characterize the relationship between geotechnical parameter field distributions and the factor of safety (FOS) using a multi-input and multi-stream Convolutional Neural Network (CNN). This meta-model enables efficient reliability analyses for colluvial slopes with highly spatially varying geotechnical parameters. The effectiveness of the proposed method is demonstrated by a typical highway colluvial slope case. The case study further reveals a linear correlation between the deterministic and reliability analysis-based ratios of safety margins. This finding suggests that both deterministic and reliability analysis outcomes can provide quantitative bases for stability assessment.</p>