As the COVID-19 pandemic continues to evolve across the globe, a large amount of data on its epidemiology has been generated. Finding relationships and predictive factors in this data can be aided by the application of machine learning and neural networks to find complex patterns in the data more efficiently and help predict the future behavior of the disease, says GlobalData, a leading data and analytics company.
Johanna Swanson, Product Manager at GlobalData, comments: “Machine learning techniques can be applied to the vast data generated by the efforts of diverse teams to help identify who is most at risk, predict the spread of the disease, and understand the origins and movement of the virus. Additionally, these lessons can be used to help predict and model any future pandemics. This can also help determine which algorithms produce more accurate predictions based on the data, allowing researchers to generate better models for the future.”
Researchers at the Big Data Research Institute at the China Pharmaceutical University showed that using a machine learning algorithm, least absolute shrinkage and selection operator (LASSO), allowed them to improve their prediction accuracy and interpretability of their statistical model for factors associated with COVID-19 risk. This machine learning model helps to select the most critical factor variables that will be used in the model, thus improving the predictive ability.
Swanson continues, “The results from this yet-to-be-peer-reviewed prediction showed that the unitary state system was positively associated with increased COVID-19 cases and deaths, as opposed to the less centralized federal system. This seems counterintuitive to what would be expected, as a centralization system would be able to concentrate resources more efficiently. This could indicate the ability to tailor responses at a more local level is more efficient at fighting the pandemic. This insight could help prevent the migration and spread of COVID-19.”
Another set of researchers, at the Earth Observation Research and Innovation Centre at the University of Energy and Natural Resources, used a One-Dimensional (1D) Convolution Neural Network (CNN) to classify and predict time series data of confirmed COVID-19 cases for all reporting countries and territories for the pandemic. The 1D-CNN predictive algorithm allowed the discovery of hidden and underlying patterns without the biases that humans may impose on the data.