Active travel is a major form of physical activity for older adults as well as an important part of their daily transport. However, the non-linear relationship between their active travel and the built environment has been insufficiently revealed. In this study, by utilizing comprehensive travel survey data in Chengdu and multi-source big data, an interpretable machine learning framework (random forest & SHAP model) was constructed to systematically investigate the non-linear impact of community-level built environment factors on older adults’ active travel propensity. The results showed that accessibility to health facilities, the sidewalk index, and the green view index are the most important three factors of built environment affecting older adults’ active travel. Various elements of built environment had complex non-linear relationship with their active travel propensity of older adults, with a significant threshold effect. The effect of built environment indicators such as accessibility to health facilities and the green view index was asymmetric. Specifically, the inhibitory effect of low values far outweighed the promoting effect of high values. This study laid a theoretical foundation and provided scientific support to develop refined intervention strategies aiming at age-friendly cities.