Table 6.

Extracted patterns showing how combinations of behavioral features selected by the machine learning algorithm are associated with high loneliness and decreased loneliness.

Pattern (features categorized into low, moderate, and high)	Postloneliness
Frequency of first screen unlock between 1 and 2 pm [weekend] = low and frequency of last screen lock between 10 and 11 am [morning] = low and time spent off campus [evening] [weekend] = low and max length of time spent at social-event houses [evening][night] [weekday] = low	Postloneliness = high loneliness Support = 17% Confidence = 92%
Number of scans of the least frequent Bluetooth device belonging to self or others [weekend, week 2017-03-08] = low and number of scans of the least frequent Bluetooth device belonging to others [morning, weekend, week 2017-05-03] = high and last screen lock between 10 and 11 am [morning, weekday, week 2017-04-19] = low and Last screen lock between 2 and 3 pm [afternoon, weekday, week 2017-02-01] [week 2017-03-15] [weekday, week 2017-03-15] = low and time at local cluster 3 [afternoon] [weekend, week 2017-03-08] = low and last screen on between 3 and 4 pm [week 2017-02-01] = low and first screen on between 3 and 4 am [night] [weekday, half semester 2017-03-01] = low	Postloneliness = high loneliness Support = 30% Confidence = 90%
Min length of phone usage [weekday] = low and min steps in active bouts [night, weekend] = low and min length of sedentary bouts [night, weekend] = low	Change = decreased loneliness Support = 31% Confidence = 92%
Last screen unlock between 2 and 3 pm [afternoon, week 2017-03-29] = low and first screen on between 5 and 6 am [weekday, week 2017-03-15] = low and last screen unlock between 1 and 2 am [half semester 2017-03-01] = low and min length of sedentary bouts [morning, weekday] = low and first screen unlock between 5 and 6 pm [week 2017-02-08] = low and minimum length of sleep duration [weekend, half semester 2017-03-01] = low	Change = decreased loneliness Support = 50% Confidence = 90%