*5.2. Laboratory Activities*

Using our framework described in Section 3, we collected data for eight different activities that would be typically performed by a technical support staff member. Table 2 illustrates the relationship between area sectors (shown in Figure 5) and activities that can take place inside each sector.


**Table 2.** Activity codes.

This mapping is based on the layout of the experimental area and provides an increased level of realism to our experimental process. For example, refilling printer cartridges takes place in Sectors 1 and 2, since this is the location of the two printers, while the scanning activity only takes place in Sector 1, as this is the location of the barcode scanner. We should also note that each activity was performed in different locations within the same sector, among participants and repetitions. For example, the network switch during the patching activity was positioned in various locations along the benches inside Sectors 2, 3 and 4. Figure 6 illustrates the activities being performed by a participant, while a detailed description of each of the activities is given below:


Our analysis focuses on recognising activities that a technical support staff member would perform and how this process can be enhanced by location information. There are, however, other activities that the participants can perform before or after they engage in one of the activities we described above. As the set of these activities depends on the context and the environment in which the system operates, we would expect that inside a computer laboratory, a participant could also be walking, standing still, sitting on a chair, etc. Our system can be adapted in order to address this. One approach we can adopt is to expand our training dataset to include a wider range of activities. This would result in a higher number of classes in our multiclass classification problem. Another approach is the inclusion of the null class, which is formed by activities that have similar patterns, but are irrelevant with the application in question. However, since in theory there is an infinite number of arbitrary activities that can belong to the null class, modelling it is particularly difficult [43].

(**c**) (**d**)

(**e**) (**f**)

**Figure 6.** The activities performed by the participants in our laboratory. (**a**) Typing; (**b**) servicing; (**c**) scanning; (**d**) relocating; (**e**) patching; (**f**) installing; (**g**) assembling; (**h**) refilling.

The data collection was carried out by using our mobile application in training mode. Data coming from the smart watch and the BLE beacons were logged by the mobile application. Furthermore, when participants were performing activities, they were only given the required basic information to minimise the amount of external influence on the participant. This allowed us to perform the activities under a more naturalistic setting, closer to real-life conditions.

Each of the aforementioned activities was performed for a time between 170 s and 180 s by three different participants, while two out of three participants repeated the activities one more time. This resulted in a total dataset duration of about 290 min.
