Table 4.

Accuracy and selected features for individual models.

Patient	Number of days	Overall accuracy (%)	Accuracy with selected features (%)	Selected features included in the classification
P1	7	85.7	100	Duration of outgoing calls, number of unique phone correspondents, and number of outgoing calls
P2	27	92	96	Changes in activity, app use time, maximum magnitude of accelerometer, and SDa of accelerometer
P3	22	92.8	92.8	Maximum magnitude of accelerometer, minimum magnitude of accelerometer, minutes sedentary, and minutes lightly active
P4	15	93.3	86.6	Most common activity, number of app changes, maximum magnitude of accelerometer, and number of awakenings
P5	28	96.1	88.5	App use time, number of app changes, and duration of outgoing calls
P6	14	92.3	100	App use time, most frequent contact phone, number of incoming calls, steps, distance, floors, minutes lightly active, and minutes asleep
P7	16	78.6	85.7	Minutes awake, maximum number of steps, number of sedentary bouts, and average number of steps
P8	16	87.5	87.5	Number of apps per minute, maximum magnitude of acceleration, number of awakenings, and average length of sedentary bouts
P9	26	85	95	Number of activities, app use time, and duration of outgoing calls
P10	35	88.9	88.9	App use time, maximum magnitude of acceleration, mean magnitude of acceleration, number of steps, and minutes lightly active
P11	16	100	100	App use time, number of app changes, and maximum magnitude of acceleration
P12	23	78.1	84.4	Location entropy, number of unique locations, time in most frequent place, travel distance, and number of steps
P13	29	84.4	91.9	Number of activity changes, maximum magnitude of accelerometer, time in most frequent place, duration of interaction with phone, and Fitbit steps
P14	21	82.6	78.3	Time in most frequent locations and average length of sedentary bouts
Average	21	88.4	91.1	-

^aSD: standard deviation.