Table 2.

Overview Feature Groups

Name	Description	Key Aspects	#
Simple Questionnaire	Primary symptom scores (WCS)58 at screening and baseline	Assumed low predictive power27,30,32,57, available before intervention start	2
Extended Questionnaire	Variety of self-report questionnaires incl. WCS58 and further eating disorder67,68, depression69, and anxiety70 symptoms and behavior patterns, personality71, self-regulation72 and self-esteem scores73, psychiatric and weight loss history, alcohol use74, socio-demographic information, and user expectations.	Assumed low predictive power27,30,32,57, theoretically available before intervention start but with high time-invest from users	51
Simple User Online Behavior	Sum of logins per day of the first week	Assumed high predictive power28,30, very simple to obtain	7
Selected User Online Behavior	Single aggregation for the first week of time to complete sessions, seconds spent, number of logins, number and length of answers, diary entries and messages to coaches and group chat	Assumed high predictive power12,13,27,31 with effort into researching and choosing most promising options and aggregation measures	13
Extended User Online Behavior	Variables from log files aggregated per day of the first week, incl. sessions completed, seconds spent, log ins, time spent in beginning/mid/end of the week and morning/day/evening, session completion, count and number of characters of diaries, group, and coach messages, exercises, answers to the sixteen most common closed questions as mean, min and max	High predictive power but possible loss due to complexity12,31,38, automatically collected during first week of intervention with limited time invest	129
Mixed Features	Extended questionnaire + selected user online behavior	Mixed, with reported increase of predictive value27	64