Richter C, et al. - Researchers sought to generate and examine a data-driven framework (feature generation based on no expert or prior knowledge) wherein only biomechanical data was used for differentiating movement patterns from anterior cruciate ligament (ACL) operated class (ACL^OP), from a limb contralateral to an ACL operated limb (ACL^{NO OP}) and a non-injured control limbs (NORM). A cohort of 156 males recovering from ACL reconstruction was used as the rehabilitating cohort (ACL reconstruction should have altered kinematics and neuromuscular properties within the operated knee) and a cohort of healthy males was used as control / normal group [62 subjects (124 NORM limbs)]. Findings illustrate that the predictive accuracy of biomechanical data is high (≈ 70%) if a movement was performed by a limb following ACL reconstruction, the contralateral limb and a limb of a healthy control group. Findings suggest that most of the information content is contained in few features, that symmetry measures are important, that understanding what a classification/prediction model has learned is significant, that different movement characteristics are captured by different exercises and that a ‘true’ normative movement pattern is not utilized by all subjects within a normative cohort.