**5. Discussion**

From the MMN results, in-ear EEG signal was verified to be highly correlated to the nearby T7 and T8 scalp EEG signals (correlation between T8 and in-ear EEG was approximately 0.853). This was expected as the 10–20 system scalp positions of T7 and T8 are just above left and right ears, respectively. They are in close proximity to ear canals. The results also correspond to the finding in previous work [17,37].

DEAP data analysis results show that using single electrode at T7 or T8 could achieve valence and arousal classification accuracies above 70 percent. This is comparable to classification accuracies obtained from using multiple EEG electrodes [7]. The results sugges<sup>t</sup> that T7 and T8 could achieve a satisfactory emotion classification level.

The results from 4.3 show that emotion classification accuracy from in-ear EEG was comparable to that of T7 and T8 (71.07% and 69.85% for valence, and 72.89% and 78.7% for arousal). The four emotion classification and arousal accuracies of in-ear EEG were slightly lower than those of T7 and T8 (53.72% and 58.12%). The valence classification accuracy was almost equal.

Furthermore, the di fferences in accuracies in emotion, valence and arousal classifications between the in-ear EEG, and T7 and T8, are not statistically significant (*p*-values = 0.74, 0.99, and 0.65, respectively). Hence, an in-ear EEG is considered comparable to T7 and T8 in emotion classification.

From the above findings, in-ear EEG was found to be highly correlated to T7 and T8. Their emotion classification results are also compatible. Hence, in-ear EEG could be considered as an alternative to scalp EEG in positions close to the ears.

In terms of wearability, in-ear EEG could be set up within five minutes and could be put on by the users themselves. During experiments most subjects did not complain of being uncomfortable or being disturbed during usage. It is also una ffected by sweat, which makes it suitable for long term monitoring in a warm climate.

The additional benefits of the in-ear EEG are also in its compatibility and familiarity to users. Earplugs, earphones, and wireless handsfree earpieces have been around for many years and people are used to them. Wearing an earpiece is considered normal, so an in-ear EEG could allow the user's acceptance much easier than conventional scalp EEG headsets. Another benefit of using in-ear EEG is the signal obtained has less artifacts from electrode movement compared to conventional scalp EEG. Scalp EEG headsets are susceptible to artifacts from the user's movement, because contacts between the scalp and electrodes could easily become loose. With an in-ear EEG that fits tightly in the ear canal, body movement causes significantly less artifacts caused by loose contact between electrode and skin [30].

Compared to conventional scalp versions, the in-ear EEG is only a single channel device, with a similar signal to T7 and T8 scalp position near the ears. That limits in-ear EEG usage. Some EEG applications are not viable, such as for attention monitoring to measure the EEG from the frontal lobe [51]. Though this has never been investigated, in-ear EEG is not expected to achieve good accuracy in attention monitoring.

A higher number of EEG channels could achieve higher accuracy in emotion classification [7], so it is a valid point to consider adding channels to the in-ear EEG. This could be done by adding more electrodes to the same earbud or wearing two in-ear EEGs on both ears. The former approach was developed in [17] with the use of a custom made earmold which is similar to the one used in a hearing aid. However, earmolds are much more costly than the generic earbuds used in this work, so additional signals would be gained at much higher costs. Furthermore, due to limited space in an ear canal, two electrodes placed there would be close together, hence similar signals are expected to be measured. The latter approach of wearing two in-ear EEGs on both ears is an alternative. It is probable that emotion classification accuracy would improve. The trade-o ff here is practicality for long term usage. A user who wears in-ear EEG on both ears will not be able to hear well, since both ear canals are blocked. Earbud redesign is needed to provide a gap in the middle to let sound through the ear canal.

Despite its potential, the in-ear EEG monitoring device would need to be further developed to be more practical. An additional feature required is wireless connectivity, possibly via Bluetooth. This would make it more convenient to use without cumbersome wires. However, the challenge is in the integrated circuit design, which needs to be able to fit into an ear canal. This point was also raised in [31].
