Predicting Sarcopenia of Female Elderly from Physical Activity Performance Measurement Using Machine Learning Classifiers

Jeong Bae Ko,1,* Kwang Bok Kim,1,* Young Sub Shin,1 Hun Han,1 Sang Kuy Han,2 Duk Young Jung,3 Jae Soo Hong1 1Digital Health Care R&D Department, Korea Institute of Industrial Technology, Cheonan, Chuncheongnam-do, South Korea; 2Robotics R&D Department, Korea Institute of Industrial Technology, Ansan, Gyeonggi-do, South Korea; 3Seongnam Senior Experience Complex, Eulji University, Seongnam, Gyeonggi-do, South Korea*These authors contributed equally to this workCorrespondence: Jae Soo HongDigital Health Care R&D Department, Korea Institute of Industrial Technology, Cheonan, Chuncheongnam-do, South KoreaTel +82 41-589-8412Fax +82 41-589-8413Email jshong94@kitech.re.krDuk Young JungSeongnam Senior Experience Complex, Eulji University, Seongnam, Gyeonggi-do, South KoreaTel +82 41-589-8412Email dyjung@eulji.ac.krPurpose: Sarcopenia is a symptom in which muscle mass decreases due to decreasing in the number of muscle fibers and muscle cross-sectional area as aging. This study aimed to develop a machine learning classification model for predicting sarcopenia through a inertial measurement unit (IMU)-based physical performance measurement data of female elderly.Patients and Methods: Seventy-eight female subjects from an elderly population (aged: 78.8± 5.7 years) volunteered to participate in this study. To evaluate the physical performance of the elderly, the experiment conducted timed-up-and-go test (TUG) and 6-minute walk test (6mWT) with worn a single IMU. Based on literature review, 132 features were extracted from collected data. Feature selection was performed through the Kruskal–Wallis test, and features datasets were constructed according to feature selection. Three major machine learning-based classification algorithms classified the sarcopenia group in each dataset, and the performance of classification models was compared.Results: As a result of comparing the classification model performance for sarcopenia prediction, the k-nearest neighborhood algorithm (kNN) classification model using 40 major features of TUG and 6mWT showed the best performance at 88%.Conclusion: This study can be used as a basic research for the development of self-monitoring technology for sarcopenia.Keywords: sarcopenia, physical activity, machine learning, inertial measurement unit