Suraj Ramchand

An Ensemble LSTM for Rare Event Detection

Abstract

In the complex study of rare diseases, it is crucial to find distinctive clinical features that allow for accurate or possibly early diagnosis. However, as suggested in the class of such conditions, its rarity is a contributing factor of it being understudied. Fabry is a type of rare disease that has a broad range of phenotypes, making it increasingly difficult to detect and diagnose. The consequences of which are dire where Fabry patients could experience premature death almost 20 years in advance. Although the condition in focus was initially Fabry, the lack of data pushed this study towards the use of Sepsis data which is similar in nature. This study explores the use of clinically sound methods of data imputation for datasets with temporal nature and have large amounts of missing data. Additionally, this study reports the features selected by the machine learning algorithms. Gleaning from the knowledge from past studies, this study shows the differences in using data that was imputed with the best clinical judgement and compares it. The findings of this study is purposed as a foundation for future work on the use of artificial intelligence in rare event detection models and temporal datasets.