Naïve Bayes versus Support Vector Machines: A comparison of two algorithmic approaches for estimating cause of death from free-text

Determining cause of death is a basic role of vital registration. Unfortunately, most deaths globally are not registered at all or without an accurate cause of death. Verbal autopsy (VA) is an alternative mechanism to estimate cause of death. However, responding to the VA questionnaire often takes a significant period of time to complete and can prove challenging as it overlaps with the mourning period. This thesis is to explore if machine learning algorithms primarily naive Bayes and SVMs can be used to determine cause of death from free-text responses from a verbal autopsy questionnaire.