Background and aims: Prediabetes is associated with an elevated risk of diabetes, cardiovascular, and microvascular complications, with highly variable progression rates. Accurately predicting progression risks and choosing effective interventions for prediabetes are crucial. We previously developed a machine learning model (ML-PR) to predict diabetes progression in prediabetic individuals within a Chinese cohort, facilitating risk assessment and guiding lifestyle and pioglitazone intervention decisions. This study aims to externally validate the ML-PR model for prediabetes risk stratification and investigate the variation in intervention effectiveness across different subgroups. Materials and methods: The prediabetes patients from Diabetes Prevention Program (DPP, a landmark trial to assess the effect of placebo, lifestyle, and metformin for up to 3 years) (n=3665) and DPP Outcomes Study (DPPOS, an extended follow-up of DPP for 15 years) (n= 2607) were included in our study. Patients were divided into tertiles as low-, medium-, and high-risk groups based on their baseline ML-PR scores. The effects of interventions on diabetes progression within subgroups were calculated by the Cox regression model and the group-by-treatment interaction was assessed by Wald test. Results: In the placebo arm of the DPP and DPPOS cohort, ML-PR predicted 1-year,3-year, and 15-year diabetes progression with Harrell's C-index of 0.78 (95% confidence interval [CI], 0.73-0.83),0.74 (95% CI, 0.71-0.78) and 0.67(95%CI,0.65-0.69), respectively. In the DPP and DPPOS studies, the high-risk group showed a significantly higher incidence of diabetes at the 1-, 3-, and 15-year follow-ups, along with more cardiovascular and microvascular events at year 15 compared to the low- and medium-risk groups (p<0.001 for all endpoints). In DPP, the hazard ratio [HR] (95%CI) of lifestyle intervention versus placebo intervention in diabetes progression was: 0.471(0.286,0.775), 0.615(0.416,0.909), and 0.372(0.287,0.482) in low-, medium- and high-risk groups (p for interaction=0.041). The corresponding figure was similar in DPPOS (p for interaction<0.001). A significant group-by-intervention interaction of metformin versus placebo intervention was observed in the DPP cohort (HR (95%CI) 0.782(0.503,1.217) in low-risk group,0.922(0.649,1.309) in medium-risk group and 0.582(0.462,0.733) in high-risk group, p for interaction=0.028), but not in the DPPOS cohort (p for interaction=0.147). Conclusion: The ML-PR model effectively predicts diabetes progression risk in prediabetic patients in large-scale US clinical trials. Our study suggests prioritizing high-risk patients for lifestyle intervention, with metformin as a potential short-term intervention. Our results demonstrate the ML-PR model's utility, emphasize the importance of tailored interventions and offer evidence for targeted clinical strategies.