On the rise of fintech development on online credit, there are numerous efforts to enhance the performance of the platforms' default risk model. One of the important criteria is the income from borrowers. Relying on detailed data from a major European marketplace lending platform, we infer borrowers' income stability and use this information to improve credit risk model performance. Applying machine learning techniques, we find that the income stability measure is negatively related to default probability. It serves a significant role to improve the performance of the benchmark model used by the platform: the AUC increased from 0.69 to 0.73 after including the income stability measure. The results hold in out-of-sample test and when using more precise sub-sample.