This article has been accepted for publication by IEEE BSN 2019 conference, but has not been fully edited. Content may change in the published version of this paper. DOI will be updated once available Developing Computational Models for Personalized ACL Injury Classification Varun Mandalapu Department of Information Systems University of Maryland, Baltimore County Baltimore, Maryland varunm1@umbc.edu Nutta Homdee Department of Electrical Engineering University of Virginia Charlottesville, Virginia nh4ar@virginia.edu Joseph M. Hart Department of Kinesiology University of Virginia Charlottesville, Virginia joehart@virginia.edu John Lach Department of Electrical Engineering University of Virginia Charlottesville, Virginia jcl7d@virginia.edu Stephan Bodkin Department of Kinesiology University of Virginia Charlottesville, Virginia sgb3d@virginia.edu Jiaqi Gong Department of Information Systems University of Maryland, Baltimore County Baltimore, Maryland jgong@umbc.edu Abstract— With the advances in wearable sensor technology, the use of inertial body sensors in the field of Medicine and Healthcare increased drastically. Researchers found that gait data is useful for identifying various motion impairments. Current research in gait analysis is incorporating the features extracted from video data which is hard to analyze and needs expensive video capture equipment to collect data in slow motion. In this study, we utilize the ability of inertial body sensors to capture gait features of individuals with Anterior Cruciate Ligament (ACL) injury. This paper also leverages the causality analysis method to find the coordination between different features of gait data. This study was applied to 131 individuals in which 109 have ACL injury. In this work, we incorporate the gait assessment technique which uses causality analysis and then predict various classification of individuals based on health condition, impacted limb prediction and performance of various machine learning algorithms. Keywords—ACL Injury, Gait Analysis, Machine Learning I. INTRODUCTION ACL injuries are common in athletes during their participation in sports. These injuries are not just prevalent in athletes, statistics show that 7 to 8 individuals out of 10,000 encounters this in general population but in sports population, this is as high as 5 to 85 individuals per 1000 [1]. Studies show that the possibility of second-time injury after the first one is 15 times greater than the uninjured individual [1]. Studies in kinesiology showed that athletes who recover poorly will have asymmetries in their inter limb correlation in their walk. These demonstrate the relationship between gait biomechanics and interlimb correlation. This will also provide physicians to identify the chances of the second incidence. Gait assessment can be done using various available methods like video annotations, manual observation and with the inertial body sensors. These sensors will demonstrate the patterns in their gait data and detect gait cycles. The data from the gait phase decomposition corresponding to temporal gait features are used very often for assessment [2]. These features include angles, gait speed, single stance, double stance time or various other parameters that are derived from this data. Individual motion consists of spatial-temporal data, interaction, and coordination between body parts. Human movements can be observed by focusing on a few tracking points and correlating their interactions. This correlation between body parts consists of rich information related to movements of the body which is more than the temporal evolution of body parts. ACL injuries due to knee tear will impact individuals gait and cause abnormal motion in body parts which can be observed by comprehensive gait assessment on all parts of the human body. The comprehensive gait performance can be calculated from the overall interactions od different body parts. From this assessment, we can observe the gait performance of ACL patients with respect to healthy individuals. In this study, we adopt the causality-based approach introduced by Gong et al [3] which will provide the quantification of various intercorrelations in body parts for better separation of ACL patients and healthy subjects. Also, we apply different computational methods that classify injured limb and present condition of the patient. We evaluate our model on 91 injured subjects and 13 healthy subjects. We also evaluate the algorithms on a gender basis and impacted limb. II. METHODS The focus of this section is to present the data collection, experimental protocol, data processing framework, equipment adapted for study and the analysis of the impact of mobility of sports injury. A. Data Collection In this study, we collect data from injured patients and healthy controls using inertial body sensors provided by Shimmer sensing. All the participants in the study are equipped with 5 shimmer sensors, these are placed on their ankles, wrists, and sacrum. We recorded accelerometer and