Localized Analysis Framework using Smartphones

Mobile phones have become an intrinsic part of our daily lives. Any data collected from them is especially reflective of our day-to-day reality. By using smartphone apps, researchers can now observe behavioral patterns in mobile data, all without soliciting any responses or relying on self-reports. Such behavioral data includes the frequency and length of phone calls, the contents of text messages, and users' physical locations. However, this immediately brings up privacy issues. Many people do not want others to harvest their mobile phone's private data. To utilize mobile data without invading privacy, I propose a localized analysis framework using smartphones. In this approach, researchers do not directly collect raw data. Instead, software on each individual's smartphone accesses the raw data locally stored in the device, and only the processed results are sent to the researchers. Researchers then analyze the aggregated individual results and find meanings at the whole sample level.
The current study aims to demonstrate utilities of the proposed localized analysis framework using smartphones with an emphasis on exploratory analysis using machine learning. While there are many possible methods to employ within this framework, in this study I adopt support vector machines (SVMs) for localized analysis at an individual level and visually inspect the aggregated patterns at the whole whole sample level. The simulation study results suggest that the proposed framework can perform as well as a conventional data collection and analysis paradigm. In this study, the aggregation of multiple individual models recovered the underlying true patterns as well as the single model built with the whole data. The localized analysis framework opens a new way of treating data to protect user privacy.