Data analytics is one of the key components in the research process of REACH because it provides actionable insights from patients’ data. Deliverable D10 extends the developments for machine learning and data acquisition for REACH done in the previous Deliverable D11b. REACH proposed and implement two different machine learning approaches that model patients’ behavior to (1) understand the different ways people respond to the same intervention and (2) detect when people’s regular behavior has changed. The latter is useful because we could determine the subpopulation for which the intervention is effective and give the intervention only to those people. Also, the insights from behavior trend clustering allow the patient to see how similar people to him or her change their behavior and learn from the responders how to improve himself or herself. The former is useful because the change in daily routines could represent a risk for the patient’s heath, thus, healthcare personnel can be alerted on time about these changes.
Both machine learning approaches work with time series data and extract behavioral patterns in an automated way. This deliverable describes the whole pipeline used in our approach, from data pre-processing to evaluation and benchmarking. The proposed models are trained and evaluated on data provided in the context of Touchpoint 2 and Touchpoint 3.
Finally, in deliverables D12 and D13 we report how we developed the machine learning approaches further and integrated them into a behavior recommender system.