Patients who experience upper-limb paralysis after stroke require continual rehabilitation. Rehabilitation must be evaluated for appropriate treatment adjustment; such evaluation can be performed using inertial measurement units (IMUs) instead of standard scales or subjective evaluations. However, IMUs produce large quantities of discretized data, and using these data directly is challenging. In this study, B-splines were used to estimate IMU trajectory data for objective evaluations of hand function and stability by using machine learning classifiers and mathematical indices. IMU trajectory data from a 2018 study on upper-limb rehabilitation were used to validate the proposed method. Features extracted from B -spline trajectories could be used to classify individuals in the 2018 study with high accuracy, and the proposed indices revealed differences between these groups. Compared with conventional rehabilitation evaluation methods, the proposed method is more objective and effective.