Bumblebee Your Way to Recovery: Transforming The Approach to Detection of Mental Health Relapses

In this paper, we present our solution to ICASSP 2024, task GC-2: The 2nd e-Prevention challenge: Psychotic and Non-Psychotic Relapse Detection using Wearable-Based Digital Phenotyping [1], First Track – detection of non-psychotic relapses, containing mainly depression episodes. The submitted solution consists of an autoencoder and an outlier detector, trained on algorithmically processed night-time data. On the test set, we achieve an average between AUROC and AUPRC of 66.56%, placing us first in the challenge.