Fusing Location Data for Depression Prediction

Automatic depression prediction using Internet traffic characteristics on smartphones

“…In addition, adding volume-based features to the combination of aggregate usage features and category-based features does not lead to further improvement in prediction results. The best correlation results are 0.54 and 0.39 for the iOS and Android datasets, respectively, which are comparable to the results when using location data collected on the phones [17,57].…”

“…Overall, the usage based features that we propose are more effective in predicting depression than volume-based features. When combining the two types of usage based features, the resulting F 1 score can be as high as 0.80, comparable to that when using other types of sensing data (e.g., when using location data [17,57]). Our results demonstrate that Internet usage data is a valuable source of data for depression screening.…”

“…Farhan et al [17] found that the features extracted from the location data are complementary to PHQ-9 scores and can predict depression with good accuracy. Yue et al [57] investigated fusing two types of location data (GPS and WiFi association data) collected from smartphones for depression detection. Lu et al [31] developed a heterogeneous multi-task learning approach for analyzing location data collected over multiple smartphone platforms.…”

Investigating the Relationships between Mobility Behaviours and Indicators of Subjective Well–Being Using Smartphone–Based Experience Sampling and GPS Tracking

“…a less regular 24-hour rhythm; Saeb et al, 2016;, (ii) lower normalized entropy (i.e. more variability of time spent at favourite locations; Farhan et al, 2016;Saeb et al, 2017Saeb et al, , 2016Saeb, Zhang, Kwasny, et al, 2015;Yue et al, 2018), (iii) lower location variance (Saeb et al, 2016;Saeb, Zhang, Kwasny, et al, 2015;Yue et al, 2018), (iv) longer stays at home (Farhan et al, 2016;Howe, Ghandeharioun, & Pedrelli, 2017;Saeb, Zhang, Kwasny, et al, 2015;Yue et al, 2018), (v) total distance covered (Ben-Zeev, Scherer, Wang, Xie, & Campbell, 2015), (vi) smaller number of location clusters visited (Farhan et al, 2016), and (vii) less transition time (i.e. percentage of time spent in a nonstationary state per day; Howe et al, 2017).…”

“…Although all of these indicators tap into the broader construct of subjective well-being, it is possible that travelling large distances is associated with higher levels of energy, but also with higher levels of stress. By measuring a wide set of indicators simultaneously, we are able to test for differential effects and investigate whether various indicators (Canzian & Musolesi, 2015) Location variance À (Saeb et al, 2016;Saeb, Zhang, Kwasny, et al, 2015;Yue et al, 2018) Standard deviation of longitude and latitude + (Ghandeharioun et al, 2017) Radius NA (Pratap et al, 2019) Speed À (Yue et al, 2018) Irregularity Circadian movement À (Saeb et al, 2016;Saeb, Zhang, Kwasny, et al, 2015 Farhan et al, 2016;Wang et al, 2014;Yue et al, 2018) Entropy Home stay + (Farhan et al, 2016;Howe et al, 2017;Saeb, Zhang, Kwasny, et al, 2015;Yue et al, 2018) Home stay À (Chow et al, 2017) Entropy À (Farhan et al, 2016;Saeb et al, 2016;Saeb, Zhang, Kwasny, et al, 2015;Yue et al, 2018) Multiple features in prediction Multiple NA (Mehrotra & Musolesi, 2018;Saeb et al, 2017;Wahle, Kowatsch, Fleisch, Rufer, & Weidt, 2016;Ware et al, 2020;Xu et al, 2019;Zakaria et al, 2019) Loneliness Distance Distance travelled À (Wang et al, 2014)…”