Leveraging Collaborative-Filtering for Personalized Behavior Modeling

The prevalence of mobile phones and wearable devices enables the passive capturing modeling human behavior at an unprecedented resolution scale. Past research has demonstrated capability sensing to model aspects physical health, mental education, work performance, etc. However, most algorithms models proposed in previous follow a one-size-fits-all (i.e., population modeling) approach that looks for common behaviors amongst all users, disregarding fact individuals can behave very differently, resulting reduced performance. Further, black-box are often used do not allow interpretability understanding. We present new method address problems personalized classification interpretability, apply it depression detection among college students. Inspired by idea collaborative-filtering, our is type memory-based learning algorithm. It leverages relevance mobile-sensed features calculate weights, which impute missing data select according specific goal (e.g., whether student depressive symptoms) different time epochs, i.e., times day days week. then compiles from epochs using majority voting obtain final prediction. algorithm on dataset collected first-year students with low data-missing rates show outperforms state-of-the-art machine 5.1% accuracy 5.5% F1 score. further verify pipeline-level generalizability achieving similar results second dataset, average improvement 3.4% across performance metrics. Beyond better novel able generate interpretations each individual. These supported existing depression-related literature potentially inspire automated intervention design future.