Empowering decision support in healthcare with AI

Abstract

Current contact tracing techniques deployed for the COVID-19 pandemic has been useful in curbing the spread of the virus. However, many of these techniques comes with their drawbacks such as incorrect identification of exposed individuals and privacy vulnerabilities. These techniques mainly adapt from the PEPP-PT or DP-3T protocols. In this project, we introduced a contact tracing system architecture to tackle the privacy issues surrounding contact tracing and explored the use of sequence embedding algorithms to embed trajectory sequences of individuals to aid contact tracing efforts. We adapt the 2 sequence embedding algorithms namely, Sequence Graph Transform and Sqn2Vec, to a contact tracing use case. Despite tuning the embedding algorithms extensively, the project was unable to achieve promising results. From our results, we speculate that sequence embedding algorithms may not be effective for our use case, because they generate embeddings based on subsequence patterns.