抄録
Healthcare data is rapidly increasing as healthcare smart devices are developed. However, statistical analysis of healthcare big data is difficult due to heterogeneity in data quality and missing data. For such data, data mining methods such as clustering and dimension reduction may be useful as their preliminary analysis. In this paper, we show two illustrative applications; one is a principal component analysis of body composition data and the other is a latent topic analysis of hourly step-count dataset recorded by physical activity monitors. From those analyses, we proposed independent indices of body size and hidden obesity and extracted some diurnal patterns in ambulatory activities.
