© 2019 IEEE. The aim of this study was to obtain objective epidemiological sleep knowledge to reveal habitual sleep characteristics in the Japanese population, such as age and gender effects, using the large-scale 24-hours Holter recording database. This database includes trunk triaxial-acceleration data for more than 80,000 individuals, collected by an accelerometer mounted on a Holter monitoring system. In this study, we first constructed a sleep-wake classifier for trunk acceleration data; 24 hours-simultaneous measurements of wrist activity data by AMI actigraph devices and acceleration data by Holter monitors was conducted (24 healthy adults). A support vector machine (SVM) was used to construct the classifier. The sleep-wake states evaluated by the Cole-Kripke sleep estimation algorithm from AMI actigraph data were used as supervised signals during training. The vector magnitude and trunk angles were used to form feature vectors. The resulting model could accurately classify the sleep-wake states from Holter acceleration data (accuracy 93.2 ± 4.3%, sensitivity 91.7 ± 8.3%, specificity 94.1 ± 4.9%). Further, we applied the classifier to the Holter database, obtaining objective epidemiological data of sleep habits, showing their significant dependence on age and gender.