Improving prediction accuracy of influenza-like illnesses in hospital emergency departments

Influenza poses a significant risk to public health, as evident by the 2009 H1N1 pandemic. Hospital emergency departments monitor infectious diseases such as influenza with surveillance systems based on arriving chief complaints. However, existing systems are too reliant on the completeness of data and are not acceptably accurate in a practical setting. To improve prediction accuracy, we propose a data cleaning process for data collected in hospital settings. Besides, we also propose a novel feature selection method called the Importance Contribution Index (ICI) which is based on orthogonal transformation. Various feature selection and pattern classification approaches are analyzed. The ICI and C4.5 decision tree are eventually adopted in the new surveillance system. Validation results have shown that the total accuracy has been improved by 7.1% in the enhanced system.