Vaccine Hesitancy Prediction Based on Machine Learning Algorithms

Abstract

Modern medical technology has led to the development and implementation of vaccines for various infectious or malignant diseases, offering active acquired immunity to those who receive them. Despite extensive research and validation confirming vaccine safety, a segment of the population exhibits reluctance or refusal to get vaccinated, known as vaccine hesitancy. This hesitancy posed challenges during the COVID-19 pandemic, hindering the attainment of herd immunity and contributing to a higher number of preventable deaths. It is imperative to accurately predict vaccine hesitancy by analyzing various population characteristics to distinguish between those with and without vaccine hesitancy. Subsequently, different communication strategies and incentives can be adopted. In this article, the author utilizes private data from a survey that includes demographic features about the participants, as well as their behaviors and beliefs about the COVID-19 pandemic and vaccination. To identify the most significant features, the author employed the feature importance method to eliminate features that were less influential and narrowed to the 17 most influential features. Machine learning techniques, including non-ensemble, ensemble, and neural networks, were employed to predict the data set and then calculate testing errors to compare each method on prediction accuracy. The authors concluded that neural networks have superior prediction accuracy and are mostly immune to overfitting. Although neural networks take a long time to train, the test error of less than 5.5% is significantly lower than the 6.5% average error of other methods.