|
Abstract:
|
Recent advances in mobile technology has introduced a novel type of data that has the potential to provide a unique insight into the behavior of patients with mood disorders. These data are collected passively and continuously, yet there could be unexpected interruptions to these continuous data streams due to various reasons: e.g. battery discharge, authorization failure, and not carrying the device. The same issue can affect actively collected self-ratings (e.g. mood, stress) obtained via daily ecological momentary assessment. We propose an unsupervised algorithm to identify missing or invalid observations and impute them before conducting analysis. For this purpose, we utilize a variety of smartphone usage measures such as screen unlocks and battery level variance. We will present both simulations and real data analysis to illustrate the performance of the proposed algorithm.
|
