In order to enhance the current methodology for failure risk determination of oil and gas pipelines girth welds and to improve the effectiveness in the selection of excavation points of risk girth welds for inspection, this paper proposes a comprehensive index system of risk factors of pipelines’ girth weld. This system, covering the entire lifecycle of pipeline girth welds, is established based on statistical analysis of oil and gas pipeline girth welds investigation data and pertinent research findings on the failure mechanism of pipeline girth welds. Additionally, a risk prediction model is developed based on the ensemble learning algorithm. This model, when compared to those models established using other machine learning algorithms, demonstrates superior prediction accuracy and versatility. The overall prediction accuracy of this model is reported to be 86.44%. The results signify that this model accurately identifies girth weld failures and serves as a beneficial guide for scientifically selecting unexcavated risk girth welds in engineering construction projects. Building upon the established prediction model, this paper introduces a method for analyzing the local importance of risk factors. This approach, obtaining the risk index factor weighting from a data modeling perspective, can serve as a reference point for quantitative risk assessments of girth welds, thereby significantly enhancing the digital management capability of pipeline girth welds.