Title Page
Contents
Abstract 10
Chapter 1. Introduction 11
Chapter 2. Related work 16
Chapter 3. Study design 19
3.1. Participants recruitment 20
3.2. Data collection 22
Chapter 4. System architecture 25
4.1. Smartphone application 26
4.1.1. Sensor data 26
4.1.2. Self-reports 33
4.2. Data collection privacy 35
Chapter 5. Methodology 36
5.1. Feature engineering 37
5.1.1. Feature extraction 37
5.1.2. Feature preprocessing 48
Chapter 6. Experiments and Results 54
6.0.1. Correlations between sensor features and the PHQ-9 item scores 54
6.0.2. Depression group prediction 69
Chapter 7. Discussion and limitations 71
Chapter 8. Conclusion 74
Bibliography 76
Table 4.1. Data sources with sampling rates and short descriptions. 34
Figure 3.1. Data flow between smartphone and server. 24
Figure 4.1. Screenshots from data collection application. 27
Figure 5.1. Activity recognition categories distribution. 50
Figure 5.2. Applications categories distribution. 50
Figure 5.3. EMA response time (all default answers). 53
Figure 6.1. Significant correlations (p 〈0.05), non-depressed group. 58
Figure 6.2. Significant correlations (p 〈0.05), depressed group. 62
Figure 6.3. Significant correlations (p 〈0.05), severely depressed group. 66
Figure 6.4. Depression group prediction accuracy for different window sizes. 70