Abstract

Those aged over 80 years are now the fastest growing population. With the effects of the constant green gas emissions observed in our world today, temperatures and heatwaves are likewise on the rise. These two factors are major contributors to dehydration. Consequently, more and more demand has been placed on hospitalization units and nursing homes in the past few years. One way to lessen such load is to prevent dehydration before it actually occurs saving time and money both for the healthcare system as well as the affected patient. The steps to prevent dehydration are clear and simple; following a correct drinking behaviour on daily basis. Such a process can be automated. In this paper we propose a smart system with the goal of preventing dehydration in nursing homes without placing much work on the concerned caregivers. The drinking behaviour of each patient is observed by capturing each time and how much a person drinks with the help of a sensor in their cup. Data captured is then sent to a server to be processed. Using the physical characteristics of each patient, an optimal drinking value is calculated and stored in our database. Further factors that would require the patient to drink more like heatwaves or extreme temperatures are received from an external API then further modification to the optimal drinking amount is applied. Lastly, we propose two different forecasting techniques with the goal of predicting how much the person in question would have drunk by the end of the day. A Deep-learning based forecasting method, namely Multilayer Perceptrons (MLPs), as well as a classical statistical-based method, Croston’s method, are applied to our use case. Upon forecasting, it would be known if an alarming message should be formulated and sent to a concerned caregiver, thus preventing a risk of developing dehydration before it occurs. Using a new proposed accuracy metric, Croston’s method is found to be a better fit for this particular use case.

Results/Implementation/Project Description

Conclusion

[ PDF (optional) ] 

[ Slides Kickoff/Final (optional)]