Start Date

11-8-2016

Description

Smartphones are increasingly utilized in society and enable scientists to record a wide range of behavioral and environmental information. These information, referred to as Unobtrusive Ecological Momentary Assessment Data, might support prediction procedures regarding the mood level of users and simultaneously contribute to an enhancement of therapy strategies. In this paper, we analyze how the mood level of healthy clients is affected by unobtrusive measures and how this kind of data contributes to the prediction performance of various statistical models (Bayesian methods, Lasso procedures, etc.). We conduct analyses on a non-user and a user level. We then compare the models by utilizing introduced performance measures. Our findings indicate that the prediction performance increases when considering individual users. However, the implemented models only perform slightly better than the introduced mean model. Indicated by feature selection methods, we assume that more meaningful variables regarding the outcome can potentially increase prediction performance.

Share

COinS
 
Aug 11th, 12:00 AM

How to Predict Mood? Delving into Features of Smartphone-Based Data

Smartphones are increasingly utilized in society and enable scientists to record a wide range of behavioral and environmental information. These information, referred to as Unobtrusive Ecological Momentary Assessment Data, might support prediction procedures regarding the mood level of users and simultaneously contribute to an enhancement of therapy strategies. In this paper, we analyze how the mood level of healthy clients is affected by unobtrusive measures and how this kind of data contributes to the prediction performance of various statistical models (Bayesian methods, Lasso procedures, etc.). We conduct analyses on a non-user and a user level. We then compare the models by utilizing introduced performance measures. Our findings indicate that the prediction performance increases when considering individual users. However, the implemented models only perform slightly better than the introduced mean model. Indicated by feature selection methods, we assume that more meaningful variables regarding the outcome can potentially increase prediction performance.