Ataxia is a neurological disorder that impairs movement coordination, requiring precise assessment methods to guide effective treatment strategies. Conventional approaches rely heavily on subjective clinical evaluations. This study seeks to establish an objective framework for measuring the severity of Hereditary Cerebellar Ataxia (HCA) and assessing rehabilitation outcomes using motion data. Guided by clinical scales, clinically relevant features were extracted. The study involved 69 individuals with ataxia from across Australia, divided into two groups: a rehabilitation cohort (n = 36) and a control group. Participants underwent truncal sitting and Romberg standing tests while wearing Inertial Measurement Unit (IMU) sensors on their back and chest. Ten supervised machine learning models were developed and evaluated against two clinical scales: the Berg Balance Scale (BBS) and the Function in Sitting Test (FIST). The KNeighbors regression model achieved the best performance, showing a correlation of 0.7613 (p = 0.01) with the FIST score for the sitting test (eyes closed) using data from the back sensor. Similarly, the Lasso Regression model demonstrated a correlation of 0.5641 (p = 0.04) with the BERG score during the standing test (feet apart, eyes open) based on back sensor data. Statistical significance was further validated by t-test p-values of 0.00641 and 0.013 for the sitting and standing tests, respectively. These results highlight the ability of the proposed framework to distinguish between rehabilitation and control cohorts in both sitting and standing tasks (p≤0.05).