COVID 19 post-vaccination adverse effects prediction with supervised machine learning models

Abstract

This electronic Pharmacovigilance with AI helps to trace possible adverse events among the vaccinated population. The symptom pattern discovery from the vaccinated population obtained through post-vaccination surveys provides insights for medical practitioners to study the possibility of adverse events among vulnerable populations. This work contains postvaccination survey data from an Indonesian national referral hospital with 840 instances, 6 categorical input features, and 15 binary target attributes. As there were multiple symptoms as a target, multi-target classification algorithms experimented on the dataset. The inadequate sample size resulted in poor performance of the algorithm. To improve model prediction performance, the target was converted into binary format. The population that exhibited at least one symptom is considered symptomatic by the binary classification models. The supervised machine learning model of test train split (80%- 20%) produced 89% accuracy with a decision tree classification algorithm in the classification of symptomatic or non-symptomatic patients.