Background:
Illness as a consequence of food-borne disease is a global concern (Redmond and Griffith, 2009). Taking a UK centric perspective, an increased prevalence of Listeria has been observed, that has been exclusively isolated to the over 60s population (SSRC, 2009, ACMSF, 2009). Conclusions of the Social Science Research Committee (SSRC) (2009) highlight the critical role that domestic food-safety and the kitchen environment has in this increase. However, there is limited insight into the consumer domestic food-safety practices in this vulnerable group.
Domestic food-provisioning practices are mundane and habitual, being tacitly enacted with little conscious thought by those involved (Backett-Milburn et al. 2010). Methods of enquiry are therefore required to disentangle complex interrelated kitchen behaviours, whilst being sympathetic to the sensitivities of the study cohort. Although traditional observational techniques are valuable, these ethnographic observations and extensive use of video equipment can be both daunting and demanding for older participants, and present logistical problems both technically and socially for researchers. This paper explores the use of accelerometers and activity recognition (AR) techniques for quantitatively evaluating home-based levels of kitchen activity and practice interaction in the over 60s.
Methodology
This study involved instrumenting 10 domestic kitchens with AX3 (Openmovement, 2011) sensors and a video camera which supported in-depth interviewing related to the 4 'C''s' of food safety:- cooking, cleaning, cross-contamination and cooling. The AX3 sensors developed by Ladha et al (2011) were overmolded in food safe plastic, making them suitable for use within the kitchen environment. Sensors were configured to continuously log tri-axial acceleration and strategically placed in a range of locations in the kitchen including inside the fridge, with each deployment lasting 14 days. A recorded video was used to annotate the accelerometer sensor data for events using the ELAN (LAT, 2008) tool. Once annotated, 50% of the accelerometer data was subjected to a principal component analysis (PCA) to identify classifiers, which was then subsequently used to train a K-Nearest Neighbour (KNN) algorithm. The remaining 50% of the gathered data was then passed through the KNN and the output compared to the hand ELAN based annotations. Results from the process indicate a sensitivity of 95.5%, a specificity of 100% and an accuracy of 97.7%.
This study demonstrates the successful implementation of accelerometers to reliably and unobtrusively evaluate levels of kitchen activity, without the need for constant and intensive observation. Contributions of AR as part of a toolkit of methods for understanding domestic-food safety behaviours and the merits for providing a complimentary layer of data to the traditional visual and the discursive were evaluated. Results of this study are presented drawing on the post-hoc reflections of both researchers and participants to evaluate the suitability of such technologies for this age cohort.