Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Philips Research Student project: Analysis and prediction of human activity and behavioural data in a wellbeing program Company name Location Company supervisors : : : Start date End date : : Philips Research Europe, User Experiences Eindhoven, the Netherlands dr. Steffen Pauws, dr. Jan Korst, dr.ir. Verus Pronk [email protected] 040 – 27 47663 to be agreed to be agreed Introduction The User Experiences group within Philips Research investigates the everyday use of a wearable sensor system that assesses movement and mobility in real life or free-living circumstances with minimal discomfort. The sensor system consists of a single small tri-axial accelerometer that can be worn easily on the body. The device comes with a key cord and a clip, so various ways of wearing are possible. It has been shown that there is a reliable and consistent relationship between energy expenditure due to physical activity and acceleration measured on the human body. A simple USB connection allows docking the sensor data to a PC platform. The current configuration uses a long-term ambulatory data collection method in which acceleration data are integrated over time periods of a single minute resulting in activity counts per minute. This activity count is a strong correlate of energy expenditure (i.e., calories burnt by the human body). Recordings up to three weeks are possible without recharging. In Figure 1, a plot is presented of a 24 hour recording of a person’s activity count. Figure 1. Raw 24 hour activity count data of a single person. Philips Research The sensor system constitutes an important element of an ongoing physical activity coaching intervention program aimed at increasing people’s daily life physical activity. In addition, the intervention program includes various motivation modalities, for instance, an intake assessment, a web site, regular communication and coaching facilities. At the intake and during the course of the program, various participant data are collected including personal characteristics (e.g., gender, age, health status), device diagnostics data (e.g., charging/docking times) and participants behaviour and program data (e.g., sensor system wearing data, web site consultation, coaching intervention, targets and performance). Many individual health / wellbeing programs require motivation for the participant to persevere, especially if these activities require a behaviour change with a long-term commitment. As a result, the drop-out rate is rather high in these programs. Aim The current student internship/graduation project aims at recognizing data patterns in a large data space including ambulatory activity data, participant characteristics and participant behaviour data to answer the following set of research questions (in order of priority): 1. How can we predict likely drop-outs of participants in a wellbeing coaching program from their activity data, possibly in combination with participant characteristics, device diagnostics data and participant behaviour data? It would be valuable, if we are able to detect and explain that a given participant requires more attention to help her persevere in the program, either right after the intake or during the course of program. 2. How can we determine optimal coaching intervention moments to prevent participants from dropping out of the program? It would be valuable to intervene, remind or provide the right resources or incentives to participants in time in order to facilitate a true behaviour change. 3. What part of the activity data can be typified as a baseline everyday physical activity level and what can be identified as additional physical effort? 4. How does the participant behave in the program? For instance, does she handle a sufficient level of discipline or does she behave more erratic? Above research questions can be re-phrased as data mining, pattern recognition and/or time series modeling problems. The number of solution methods (i.e., algorithms) for these problems is large. A large anonymised data set is available for research. This set includes ambulatory activity data for 84 days that is recorded from a tri-axial accelerometer worn in daily life conditions under a wellbeing coaching program. In addition, participant characteristics, device diagnostics data, website consultation data, coaching intervention/program data and device wearing data are also logged. Recruitment We are looking for university students with an interest and understanding in data mining, machine learning, pattern recognition and/or time series modelling, who want to do their internship or graduation project at Philips Research. The student needs good programming skills for realizing the project and good technical writing skills for reporting. Creativity, commitment, a proactive attitude and own initiatives are important. Students will work under guidelines applying at Philips Research in Eindhoven on a temporary contract. For students living and studying in the Netherlands, allowance is determined at 330 euro per month for an internship and 420 euro per month for a graduation project. Students do not get a contribution in travel costs. For students living and studying abroad, other rules apply. Special arrangement for summer student fellowship is also possible. For more information about the project, please contact Steffen Pauws; [email protected]; 040 – 27 47663.