**Appendix**

#### *Appendix A.1 Adapted Barcode Design and Complexity Computation*

A 'barcode' state was assigned to each second of the time series according to Table A1. The definition was modified for the type 'active' from the original design presented in [10] that applied the absolute value of acceleration. As acceleration is sensitive to the sensor wearing location, the acceleration based threshold was replaced by the 'ActiCount' in this study. The 'ActiCount' thresholds are determined based on a validation study presented in [19]. Complexity computation was according to Equation (A1), where 'nrPattern' is the total number of sub-patterns found in the 'barcode'. 'nBC' is the total number of 'barcode' states. In our analysis, nBC = 18. N is the total length of PA time series in seconds.

$$\text{Complexity} = \frac{\text{nrPattern} \ast \left(\frac{\log\_{10}(\text{nrPattern})}{\log\_{10}(\text{nBC})} + 1\right)}{\text{N}} \tag{A1}$$


**Table A1.** Definition of barcode state according to PA category, intensity and duration.

*Appendix A.2 Effect of Smoothing on the Duration of Activity Bouts and Reliability of Complexity*

Figure A1 shows that, after smoothing, 90% of the sedentary bouts lasted up to 15 min, whereas only 10% of the walking bouts were longer than two and half minutes before a stop. Figure A2 compares the mean and standard deviation of daily complexity in one-week measurement before and after smoothing. Smoothing lowered the mean values and the CV of the complexity.

**Figure A1.** Comparison of cumulated distribution of walking and sedentary bouts before and after smoothing of activity classification output.

**Figure A2.** CV before and after smoothing PA time series.
