Ensuring fairness in scoring is a critical challenge in judged sports. While automated scoring systems that predict scores from performance data have emerged as a promising solution, the annotated scores used for training such models often reflect subjective biases among individual judges, leading to inconsistencies in the dataset. To address this issue, we propose a learning framework that incorporates relative ranking information between performance samples—relationships that are typically more stable across different judges. By focusing on ranking-based learning rather than direct score regression, our method aims to reduce the influence of individual scoring biases and improve model consistency. We conduct comprehensive experiments on three judged sports datasets, including a newly constructed figure skating dataset. The results show that our ranking-based learning framework significantly improves the prediction accuracy of automated scoring models compared to conventional regression-based methods.