Data Analysis and Generation in the ENVELLINT Longitudinal Study to Determine Loss of Functionality in Elderly People

This paper presents the initial data analysis and modelling for detecting health changes from data gathered on a low-cost smartphone used during normal daily activities. The work is part of the ENVELLINT project, where one of the main objectives is to explore if it is possible to evaluate the functional aspects of frailty indices automatically using smartphones. The project involves both longitudinal and cross-sectional studies involving elderly participants. In the longitudinal study a comprehensive set of sensor, application and other smartphone data is gathered over lengthy periods for each participant, together with extensive medical assessments. The purpose is to provide a comprehensive data set for investigating frailty and health changes. The larger cross-sectional study, which included only the medical assessments, was necessary to gather more medical related health and frailty data, and to balance project costs. The analysis work to date has involved data and feature engineering to identify, extract and select the most useful features. Insights are given for the potential use of the location and application usage features. A core aspect, given the expense and the limited number of participants in the longitudinal study, is to explore the use of synthetic data generation to leverage the real data from both studies. Generative Adversarial Network and Gaussian Copula models have been investigated to create a larger representative dataset of longitudinal participants. Initial results and insights show generated synthetic data that closely mirrors the real data, especially using Gaussian Copula.