**4. Conclusions**

In this paper, we have presented a solution for the detection of the level of arousal from electrodermal signals (EDA) in people through their exposure to musical stimuli. For this purpose, participants over 60 years old from the region of Murcia, Spain, were recruited to listen to a series of musical pieces similar to those performed in their youth. During the playback of the music, the EDA of the participants was continuously monitored. The EDA signals acquired during the experiment were then used, along with a SAM questionnaire filled out by the participants, to conduct a couple of studies. A first study looked at the features of EDA and their ability to check for statistically significant differences for each feature extracted. A second study used well-known classifiers to analyze the potential correlation between the objective detection of the level of excitation of processed physiological EDA signals and the level of arousal subjectively perceived by participants when answering the SAM questionnaire.

The first study was based on the analysis of the existence of some kind of statistically significant difference in the selected features. The study found a greater number of statistically significant differences in the musical genres of Flamenco and Spanish Folklore, and much less in the genre of rock/jazz, which seems reasonable in the Spanish region under consideration. One of the most important factors determining musical preferences is familiarity. In accordance with our study, becoming familiar with a particular piece of music has demonstrated to increase a subject's level of enjoyment [65–68]. This is true for emotional and autobiographical memory experiences provoked by musical stimuli [68,69]. The use of new musical stimuli allows us to control familiarity, since they are stimuli that have not been heard before. In this work the use of the same neutral melodic base in all the musical fragments (own design of the study) on different musical styles was considered. The only variation in the experiment are the musical genres, so any differences we may find must be due to the styles and not to familiarity with the musical stimulus. Considering that EDA is very sensitive to familiarity and prior exposure, the use of the procedure used in this proposal provides an important advance in music psychology research.

The second study, based on classifiers, provided information on the ability to distinguish between low and high arousal levels using both the processed EDA signals and the responses to the SAM questionnaire completed by the participants. This second study concluded that SVM, KNN and ensemble trees are classifiers that work very well in this case. Other classifiers such as linear discriminant and logistic regression did not work well for any music genre. In relation to the second study, this work has some limitations both in terms of the number of participants and the selection of the EDA signal features. In first place, a larger number of participants would be necessary to

reinforce the results obtained in this study. Second, despite the large number of features used during the machine learning process, overfitting was not detected in the experiment presented. Nonetheless, a more in-depth investigation on the reduction of the features would be of interest.

This study has another limitation that has to do with the evaluation of the participants' musical experience. Although stylistic variations of a new piece have been used, the previous exposure to the styles may not have been the same for the participants. Therefore, a system should be developed to evaluate the baseline of each participant in future studies.

The main contribution of this article has been the study of different music genres in older people to achieve a positive influence on their emotions and thus mitigate negative effects such as anxiety and depression. That contribution is based on the possibility of raising the arousal produced by memories evoked from their youth through the music heard at that time. The results of this work open the door to further studies on the fluctuations of EDA in older people with depression and/or cognitive impairment. We believe that these discoveries are expandable by developing new automated systems to help older people in their daily lives. Based on the results achieved in this experiment, we will be able to develop ambient intelligence systems to improve the quality of life and well-being of the elderly.

**Author Contributions:** Conceptualisation, A.B.-T. and A.F.-S.; methodology, R.S.-R.; software, R.S.-R.; validation, A.B.-T., A.F.-S., R.S.-R. and A.F.-C.; writing–original draft preparation, R.S.-R., A.B.-T. and A.F.-S.; writing–review and editing, A.F.-C. and J.M.L.; funding acquisition, A.F.-C. and J.M.L. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work has been supported by Spanish Ministerio de Ciencia, Innovación y Universidades, Agencia Estatal de Investigación (AEI) / European Regional Development Fund (FEDER, UE) under DPI2016-80894-R grant, and by CIBERSAM of the Instituto de Salud Carlos III. Roberto Sánchez-Reolid holds BES-2017-081958 scholarship from Spanish Ministerio de Educación y Formación Profesional.

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