**7. Conclusions**

In this study, we classified boredom using features from EEG and GSR datasets that were trained and tested by 30 models based on 19 different machine learning algorithms as an initial test for finding a suitable classification algorithm. We picked MLP, RF, and NB as the most suitable candidate algorithms. After tuning the selected algorithms' hyperparameters, we executed 1000 iterations of 10-fold cross validation with different random seed values to identify the most robust model among these. As a result, we recommended the MLP model which had a mean accuracy of 79.98% on the EEG-GSR combined dataset. Another major finding is that EEG and GSR appear to correlate with boredom, thus supporting the conclusion of Bench and Lench [34] that boredom and autonomic nervous system are linked.

Although this study produced novel contributions, there are noteworthy limitations. First, we collected physiological data from young and healthy participants. Thus, the recommended models may not be applicable to other age groups and to people with health issues. In addition, we hypothesize that emotion elicitation, and possibly also the manifestation of experienced emotions, is related to culture. We collected the data from Korean participants using non-boredom stimuli that were purposefully picked for this cultural context; therefore, the model may not be applicable to participants coming from other cultures. Regarding the protocol, we did not consider the effect of the order of showing the stimuli because the number of participants was deemed to be insufficient for dividing them into comparison groups. Finally, we used only one type of content to elicit boredom. Other types of contents may give different results about the intensity of the experienced emotion. In our future work, we aim to solve these limitations by collecting more data from a diverse group of users who are exposed to different boredom-evoking stimuli.

These results can be of use to developers building accurate affective computing systems as well as to researchers who seek to understand the physiological properties of boredom. As noted above, the current results still have limited applicability due to the experiment design that used only one type of boredom stimulus and a fairly homogeneous participant population. We plan to use diverse stimuli and extend the data collection to children, elderly people, patients suffering from different medical conditions, and participants representing other cultures to overcome these limitations.

**Author Contributions:** J.S. contributed on data collection, performed the experiments, and wrote the original draft; J.S., T.H.L., and K.S. conceived and designed the methodology; T.H.L. and K.S. supervised this study, and contributed on analysis and writing; K.S. administrated the overall project.

**Funding:** This research was supported by the MSIT (Ministry of Science and ICT), Korea, under the ITRC (Information Technology Research Center) support program (IITP-2019-2018-0-01431) supervised by the IITP (Institute for Information & Communications Technology Promotion), and also by the National Research Foundation of Korea gran<sup>t</sup> funded by the Korea governmen<sup>t</sup> (MSIT) (No. NRF-2019R1A2C1006608).

**Acknowledgments:** We would like to thank the participants who joined the experiment. Furthermore, we extend a special thanks to Byungkon Kang and Jaesik Kim at Ajou University.

**Conflicts of Interest:** The authors declare no conflict of interest.
