A TRANSFER LEARNING – BASED CLINICAL DECISION SUPPORT SYSTEM FOR EARLY YELLOW FEVER PREDICTION

Abstract

: Predicting the occurrence of yellow fever is difficult due to availability of fewer data quality, insufficient resources in he endemic region and the complex nature of vector borne diseases. Conventional methods of infectious diseases prediction such as y ellow fever rely on delays in reporting, non integration of heterogeneous data and inadequacy of simulation or computational modeling. While artificial intelligence holds promise for address challenges like this in prediction of disease , it has yet to be b roadly explored for predicting yellow fever. These shortcomings affect our capacity to predict outbreaks and take timely preventative measures. In this study we fill this gap by comparatively applying a transformer based Tab PFN Classifier (using transfer learning over tabular data) and a conventional Random Forest Classifier to build an efficient Clinical Decision Support System (CDSS) for an early warning or intervention. Using a benchmark dataset for prediction, the CDSS will comprise binary indicator of characteristic symptoms (fever, headache, hemorrhagic symptoms) for the presence of yellow fever against binary output value (yellow fever or no yellow fever case). Tab PFN was characterized and evaluated through repeatedly Stratified 5 fold cross validat ion (5 repeats), performance metrics achieved are 0.9889 (SD = 0.010) in accuracy, 0.9600 (SD = 0.029) in precision, 0.9240 (SD = 0.034) in recall, 0.9383 (SD = 0.022) in F1 score and 0.9623 (SD = 0.074) for ROC AUC. The RFC achieved an accuracy of 0.9804, 1.00 precision, 0.50 recall, 0.6667 F1 score and 0.75 ROC AUC based on 80% training and 20% testing dataset splits. Our comparative analysis shows a substantially better sensitivity and robust performance of Tab PFN Classify which is vital in a clinical s etting and where costs of missing a clinical diagnosis is high. These results demonstrate potential for using transfer learning enabled tabular models as an effective CDSS to augment prediction of yellow fever. Nevertheless, further analysis needs to be ca rried out on extensive data.

Keywords: Random Forest, Yellow Fever, Transfer Learning, Decision Support System