Dockless bike-sharing (DBS), with its advantages of flexibility, environmental friendliness, and good physical fitness, is regarded as an effective approach to relieve urban traffic issues. As such, various studies have been conducted to explore the impact of the built environment on DBS usage. However, few studies have investigated the nonlinear effects at the street level. Especially, existing studies provided no insights on the influence of traffic conditions on DBS trips. By taking the central districts of Shenzhen, China as the case, this study proposed a new method for extracting the street-level DBS trips by considering the bicycle trajectory, and charactered the street-level associated factors by using multi-source big data. Moreover, we further employed an advanced supervised machine learning approach (i.e., gradient boosting decision trees, GBDT) and the machine learning interpretation methods (i.e., relative importance and partial dependence plots) to examine the nonlinear effects of traffic conditions and built environment factors on DBS trips at the street level. The results indicated a significant positive association between vehicle flow and DBS trips. High driving and riding traffic are more likely to occur simultaneously on urban primary arterials. Furthermore, vehicle PM2.5 emissions are positively correlated with DBS trips during peak commuting hours, but negatively correlated during non-peak hours. The effective range and threshold effects of the other factors on cycling were further identified. These findings can inform scientific decisions on the improvement of non-motorized transportation systems and cycling-friendly environments in metropolitan areas.