*5.1. Participants*

We recruited 10 participants (7 males and 3 females) aged between 25 and 34 (*M* = 29.9, *SD* = 3.4). Among these participants, four were from the first study, another four were from the second study and two were newly recruited. A requirement for our participant selection was a foot size that matched the prototype. In addition, the participant had to work in an office and spent the majority of the time in a sitting posture (>70%). Apart from this, the inclusion/exclusion criteria remained similar to previous studies.

#### *5.2. Task and Procedure*

The study was conducted on a particular day that the participants expected to have some periods of acute stress. Some of the stress tasks included working for a deadline, debugging a firmware/software, having a meeting with their supervisor, writing a paper for an upcoming submission, etc. Having lunch/coffee with friends and having casual chats with friends were some activities that would supposedly relax the participants.

The study began at 09:45 a.m. local time at the participants' office space. After elaborating on the study procedures, filling consent forms and collecting demographic data, the experimenter asked participants to wear the StressFoot and E4 wristband. At 10:00 a.m., the experimenter initialised apparatuses for data collection and asked participants to fill a questionnaire, asking them to rate their current stress level, energy level and how pleasant they felt on a 7-point Likert scale. Then, participants were asked to continue their work as usual. We asked participants to fill out two forms, a calendar application reporting the type and duration of tasks performed, as well as a questionnaire. These were the only tasks we asked the participants to perform hourly over 8 h from 10:00 a.m. to 6:00 p.m. To ensure the participants would remember to report and to fill out the calendar and the form, they received reminders via visual and audio pop-ups at the end of each hour. The questionnaire asked the participants to rate their perceived stress level, energy level, as well as how pleasant they felt during last hour. Part of the questionnaire was to fill a NASA TLX, which we used to calculate the overall workload.

#### *5.3. Apparatus and Data Gathering*

Similar to the previous studies, the StressFoot prototype was used to collect foot motion and pressure data. The Empatica E4 wristband was used to collect EDA, as well as motion data from the participants.

The data gathering was thus similar to the previous study. Additionally, we collected accelerometer data from the E4 wristband to identify motion (walking) and posture (sitting/standing). We used the E4's preset sampling rate of 32 Hz. The data analysis also remained similar to the previous study. Next, we segmented the accelerometer data into non-overlapping windows. Literature suggests using a window size less than 10 s, such as 2.5 s [92] or even 1 s [93]. Through an experimentation with three users, we determined 5 s as a suitable window size for our use case. To identify motion and posture, we relied on a threshold analysis with a single feature, as suggested by Gjoreski et al. [94]. The feature used is the first derivative over the entire window, also known as an "Acceleration Vector Change" (AVC) [94]. This feature can identify: walking (AVC >= 2 ms<sup>−</sup>3), standing (AVC < 0.1 ms<sup>−</sup>3) and sitting (0.1 ms<sup>−</sup><sup>3</sup> =< AVC < 0.2 ms<sup>−</sup>3). We applied this to extract the timestamps to identify the sitting time. Using these timestamps, we selected the sitting data from the corresponding foot motion and foot pressure data from the StressFoot prototype.

Finally, the sitting data was segmented into 10 s of non-overlapping windows. Then, all features were extracted, as mentioned before, and classified using the best performing model, which was a multi-feature model (A1+B2+C3+D1). At every 10 s window, the model classifies whether a participant is stressed or relaxed. We then calculated the percentage of the number of windows that classified stressed. This provides a measurement of the duration that a participant might have been stressed. Therefore, for each hour, we calculated a ratio ( *RS*), where:

$$R\_{\mathbb{S}} = \frac{n \,\mathrm{Windows}\_{\mathrm{siting}}(Stresscd)}{n \,\mathrm{Windows}\_{\mathrm{siting}}(Total)} \tag{11}$$
