Prediction of Future Strains of Influenza A Subtypes using Machine Learning Techniques

Abstract

The influenza-A virus has always been one of the most challenging and serious health concerns among humans due to its rapidly changing evolutionary behavior. The only solution is to take vaccine every year. Due to its rapid mutation, the structure of the virus changes rapidly. Along with the antibodies, the structure of the virus is also play very important role in vaccine preparation. The proposed work aims to perform comparative analysis of tree based prediction techniques. This will help in find out a suitable model to predict the future strains of influenza-A virus for hemagglutinin(HA) and neuraminidase (NA) glycoproteins based on previous trends in flu mutations. In this study we have taken four subtypes of influenza virus A namely H1N1, H5N1, H3N2, H9N2 for HA (hemagglutinin) and NA (neuraminidase) protein sequence along with its corresponding child sequence. We have experimented on Decision Trees, Random Forests, Extra trees & Multi-output Meta estimator ensemble techniques for predicting the future sequences. It generates a predicted offspring for the specific input flu strain. The performance of the four ensemble techniques have been analyzed using three cross validation techniques viz. Holdout validation, Repeated random test-train split, K-Fold cross validation. It has been observed from results that Random forest performs better compared to Decision tree, Extra tree, and Multioutput Meta estimator techniques with respect to Mean absolute error, root mean square error and r2 score parameters.