A Machine Learning Approach for Saddle Height Classification in Cycling

Fangbo BingGuoxin ZhangLinjuan WeiMing Zhang^*

• Hong Kong Polytechnic University, Kowloon, Hong Kong, SAR China

Background: Saddle height is an important factor in bike-fitting because it correlates with cycling efficiency and the risk of injuries. Conventional approaches mostly use anthropometric parameters and joint angles as the reference of the optimal saddle height, such as greater trochanter height and knee flexion angle. However, these methods fail to consider individual dynamic differences in cycling. Objective: This study proposed a machine learning (ML) model for discriminating saddle height levels based on easily measured kinematic data. Method: Sixteen subjects participated in riding tests under three saddle height levels. The motion capture system recorded the trajectories of markers attached to the lower limbs. Features were calculated from the joint angles of the hip, knee, and ankle. The optimal feature set was selected by forward sequential feature selection. The accuracies of four ML models were compared using leave-one-subject-out cross-validation. Results: The optimal feature set contains 14 features related to the hip, knee, and ankle joint angles. The sagittal-plane knee angle was the most sensitive to the saddle height with a classification accuracy of 80%. The k-nearest neighbor model had the highest accuracy of 99.79% when using all optimal features as inputs. Conclusion: The proposed model compensates for the lack of consideration of individual dynamic variations in cycling in traditional methods, which is a more objective tool for data-driven personalization in bike fitting.