|
Purpose of the Project
This project proposes a personalised risk assessment App. The aim of the project is to create a machine learning risk assessment algorithm in a mobile app that will determine what’s normal for a user and alert a caregiver if there are changes from normality.
Description of the Project
There were three main objectives of the project: (1) to understand the inputs that are relevant to a user, e.g. traveling to town, wake/sleep times, contact with family/friends; (2) to develop an adaptive machine learning algorithm to adjust the level of risk for a particular user; (3) to build an app to capture the input data, implement the algorithm and communicate with care givers.
Data
This work is experimentally based. Using the MIT App Inventor we created a demonstration app to detect sensory interactions of the person. The testing sample consisted of a representative sample of family and friends.
We set out three key areas for investigation:
- A Baseline Survey of participant interaction and an indication of the need for assistance
- Evaluating the effectiveness of crisp boundary logic
- Evaluating the effectiveness of multivariate fuzzy logic
Conclusions
Both the crisp boundary and fuzzy logic approaches showed a close correlation between the estimated and survey results: crisp boundaries 66%; fuzzy logic 88%. The fuzzy logic approach had an average 22% performance improvement in comparison to the crisp boundary approach.
|