This paper investigates a quantization and clustering issue on human motion performance constrained by disabilities. In a longitudinal study of medical therapy on motion disorder, stages of patient disability condition change over time. We investigate four different stages of one arm constrained walking motions by restricting 0%, 10%, 16% and 22% of arm swing angles. For analysis we use One-way ANOVA and K-mean clustering to indentify the most significant features and to partition four different motion constrained groups. Our experimental result shows that all four arm constraints during walking motion are clustered with an average accuracy of 91.7% on two different feature conditions: a mixture of singular value decomposition (SVD) and power spectral density (PSD); and SVD only on selected gait cycles. The proposed method can be integrated with a ubiquitous system (using wearable sensors) for a remote distance patient monitoring system analysis.