A supervised machine learning approach for diagnosing Lassa fever and viral Hemorrhagic fever types reliant on observed signs

Abstract

Lassa hemorrhagic fever is an infectious life-threatening fever characterized by bleeding caused by the single-stranded virus of the Arenaviridae virus family transmitted to humans via contact with blood, urine, food, or household items contaminated with rodent urine and/or feces, and other body secreted fluids from an infected person with the Lassa virus. The symptoms of this disease are fever, general weakness, malaise, headache, sore throat, chest pain, nausea, vomiting, bleeding from the mouth and nose just to name a few. In 2011, the World Health Organization (WHO) declared Lassa fever as an endemic and pandemic due to the spread of the Lassa virus in West African countries such as Benin, Ghana, Guinea, Liberia, Mali, Sierra Leone, and Nigeria respectively that has caused millions of death yearly due to a lack of early diagnosis of the ailment in this region. In the recent past, several systems have been developed to diagnose this endemic disease, but they generated a lot of false-negative during testing and were unable to detect Lassa fever, its overlapping symptoms, and Viral Hemorrhagic fever types. Hence, in this paper, we proposed and simulated a model to diagnose Lassa fever, and Viral Hemorrhagic fever types using a machine learning technique called Bayesian Belief Network. The model was designed using Bayes Server and tested with data collected from a Viral Hemorrhagic Fever medical repository. The model had 100% overall prediction accuracy based on test data; with 98.73% sensitivity of Lassa fever, and 98.98 sensitivity of Viral Hemorrhagic fever types in that order.