Predicting Subjective Measures of Social Anxiety from Sparsely Collected Mobile Sensor Data
summary
Smartphone passive sensors (via MetricWire) and nightly 10 PM EMAs collect day-level features from 80 participants over five weeks to predict seven self-reported social anxiety dimensions. Four imputation methods (MICE, KNN, LOCF, Matrix Completion) and regression models (CatBoost, MERF, RF) are compared; CatBoost handles missing values best without imputation, while MERF with Matrix Completion yields the lowest errors, with imputation improving RMSE by up to 22% RMSE reduction. Little’s test returns p=0.44, supporting imputation.
AI-written summary · about
chapters (6)
- 0:00Start
- 0:37Ecological Momentary Assessment (EMA)
- 2:49Missing Data?
- 3:04Imputation Techniques
- 3:29Predictive Models
- 5:30Summary of Findings
related talks
06mPerson-Centered Predictions of Psychological Constructs with Social Media Contextualized by ...
27mHARSHA URLAM & SWETA SINGH at Cypher 2017
06mExamining the Social Context of Alcohol Drinking in Young Adults with Smartphone Sensing
07mEvaluating the Reproducibility of Physiological Stress Detection Models
06mOne More Bite? Inferring Food Consumption Level of College Students Using Smartphone Sensing and ...
06mDetecting Stressful Social Interactions Using Wearable Physiological and Inertial Sensors
21m“Trust Us”: Mobile Phone Use Patterns Can Predict Individual Trust Propensity
26mMachine Learning for Personalised Health