Abstract: This study aims at estimating the effects of road infrastructure investment on rural poverty incidence. Using the panel data of 31 provincial administrative regions in China from 2013 to 2018, we apply a Gradient Boosted Decision Tree (GBDT) model to explore the possible non-linear relationship between road infrastructure and rural poverty incidence in the perspective of the spatial-temporal heterogeneity. In our model, the rural poverty incidence was taken as a response variable, and scale, density, investment, and hardening of road infrastructure were set explanatory variables. The results show that factors with higher relative importance include the hardening township, and village roads, the length of village roads and the expressway, spatial-temporal heterogeneity, and road density. The effects of poverty alleviation of the road infrastructure overall rely on the improvement of road density and the hardening and construction of county, township, and village roads. The partial dependence plots illustrate that the length of highway and village road, road density, rural road investment, and hardening road length is positively related to poverty alleviation. These factors present significant marginal diminishing effects on both two different periods. The length of highway and village roads present threshold effects on rural poverty incidence, which means poverty alleviation occurs only when they reach a certain scale. The performance of road density, rural road investment, and hardening road length are also like the length of highway and village roads. The poverty alleviation effects will keep stable if the above indicators exceed their thresholds. These findings provide helpful references for the planning and construction of rural roads during the transition period between comprehensive poverty alleviation and rural revitalization, and they also assist to establish a persistent mechanism to monitor and alleviate the rural poverty incidence.