**Abbreviations**

The following abbreviations are used in this manuscript:


#### **Appendix A. Features from the ECG Signal**

The measurements used to characterize the ECG can be divided into two groups, those calculated in the *time domain* and those calculated in the *frequency domain*. Figure A1 shows the features extracted from the ECG signal. As we can appreciate, there are 83, 89 and 2 features for the time domain, frequency domain, and the mixed domain, respectively. That means that the total number of features calculated to characterize the ECG measurement is 174.

**Figure A1.** ECG-based feature extraction scheme.

#### *Appendix A.1. Time-Domain*

A total of five measurements were considered in the time-domain, directly or indirectly extracted from the QRS analysis:

	- **–** Deep Breathing Difference (DBD), calculated as the difference between the maximum RR and minimum RR in the window under study [81,82].
	- **–** Ratios maximum RR vs. minimum RR, that is, RRmax/RRmin and RRmin/RRmax [83].
	- **–** Logarithm of the standard deviation of RR in the window under study.
	- **–** Respiratory Sinus Arrhythmia (RSA), calculated as the quotient between the DBD and the mean value of the RR in the window under study. This measurement is related to the function of parasympathetic nervous during spontaneous ventilation [84].
	- **–** Modal Value (MV), defined as the most frequent value in the RR intervals in the window under study [40].
	- **–** Load Index (LI), based on the ratio between the number of occurrences of each Modal Value and DBD [40].
	- **–** NN50, determined as the number of successive RR interval pairs differing by more than 50 ms [40,42,48].
	- **–** pNN50, obtained dividing NN50 by the total number of RR intervals [40,42,48].

Many other measurements were found in the literature such as SDNN index, SDANN among others [34]. However, we did not use these measurements because they require at least 5 min to be calculated, since they are often calculated over a 24-h period.

### *Appendix A.2. Frequency-Domain*

The other main group of measurements are evaluated in the frequency domain, through the Discrete Fourier Transform (DFT). The main measurements taken in this part were:


The parameters taken from these spectral measurements were the SSSPs, the baseline parameter, and a set of specific parameters related to ratios between the average power for the different bands: HF/LF, LF/HF, MF/HF, (LF+MF)/HF, and HF/TF, being TF the total power in all frequencies [23,43,44,46,47,49].

#### *Appendix A.3. Mixed Domain*

There were also two parameters taken from relationships between time and frequency parameters, denoted as Coefficients of Component Variance (CCV). The CCVs considered were the CCV-LF and the CCV-HF [39], and they were calculated as the square root of LF or HF power divided by the average HR.

**Figure A2.** TEB-based feature extraction scheme for the classical set of features.

#### **Appendix B. Features from the TEB Signal**

The measurements used in order to extract the most relevant information from TEB signal follow a structure similar to the one described in the case of the ECG signal, being again divided into the time domain, frequency domain, and mixed domain features.

Figure A2 shows the features extracted from the TEB signal. There are 60 time domain features, 89 frequency domain features, and 2 mixed domain features. Therefore, the total number of features calculated to characterize the TEB signal is 151.

#### *Appendix B.1. Time Domain*


**Figure A3.** EDA-based feature extraction scheme.

#### *Appendix B.2. Frequency Domain*

The frequency features calculated from TEB are similar to those calculated from ECG. So, again the PSD of the original signal was calculated and applied to several filtered versions of the signal. Again, apart from standard parameters, power ratios were also evaluated.

#### *Appendix B.3. Mixed Domain*

The features considered are the CCV mixed-domain parameters (in this case, using the average BR to normalize the squared root of the energy).

#### **Appendix C. Features from the EDA Signal**

The structure used in order to extract the most relevant information from the EDA signal follows a structure similar to the one described in the case of the ECG signal and the TEB signal.

Figure A3 shows the measurements obtained from the EDA signal and the procedure used to extract the features. In total, we obtained 62 time domain features and 42 frequency domain features. Taking into account that the EDA was registered in both the hand and the arm (as was described above), we obtained 104 features from the arm EDA and 104 features from hand EDA.

#### *Appendix C.1. Time Domain*

We obtained four different time domain measurements:


measurement, typical parameters are extracted using the SSSPs and the baseline parameter, and also some specific parameters:


#### *Appendix C.2. Frequency Domain*

The PSDs extracted from the EDA-Original signal, the EDA-LF and the EDA-HF, were taken as spectral measurements, using a Welch's overlapped segmen<sup>t</sup> averaging estimator. The SSSPs and the baseline parameter were calculated from these measurements.
