Mobile sleep monitoring and automatic sleep scoring are two tightly linked topics, which have both seen important developments the past few years.
In this paper, we combine leading efforts from both fields, the ear-EEG sleep monitor and a state-of-the-art sleep scoring model, ‘seqsleepnet’, to investigate the upper limits of mobile sleep scoring. We manage to further improve on the state of the art in this field, and perform a detailed analysis of the influence of electrode positioning. As part of the analysis, we test 13 different electrode combinations, each chosen for being plausible setups for mobile sleep monitoring. From this, we identify a general rule of thumb: as long as a setup contains EOG information and electrode distances on the order of the width of the head, then good automatic sleep scoring is possible. By comparing the pairwise agreement between all automatic and two manual sleep scorings, we find indications that the best of the automatic scorings may be more reliable, and equally correct, as the manual scoring. Additionally, we find that the primary source for discrepancies between automatic and manual scoring is stage transitions, and that the non-REM 1 stage is considerably harder to score consistently than the other four stages.