Predicting unreliable response patterns in smartphone health surveys: A case study with the mood survey

Due to the advancement of smartphone sensing and computing capabilities, healthcare providers are increasingly relying on smartphone-based self-reported surveys to gather information regarding health and well-being for the purpose of analyzing various factors and their impact on people’s health and well-being. However, the analysis could be misleading if the collected data is not reliable, which can happen due to several reasons, including lack of attention while taking the surveys. There is a dearth of knowledge about the reliability of smartphone-based health and well-being surveys. In this work, we present a case study with 86 participants responding to the Positive and Negative Affect Schedule (PANAS) mood survey twice — first unreliably and then carefully. From our detailed analysis with a set of 10 influential features, we are able to detect an unreliable submission with an accuracy of up to 0.93 using cross-validation. Findings from this work can be taken as guidelines for future research.