UC Santa Cruz

Abstract

Data Mining for Improving Health-Care Resource Deployment

Nannan He

While the health care industry accounts for a significant large portion of the GDP, the health care system in the US are still relatively inefficient. Before cutting down unnecessary health care expenses, it is important to ensure that individuals who really need medical attention should receive it. For example, if we could predict the hospitalization period (in days) for a potential patient, then we could better predict and distribute health care resources.

In this research, we apply data mining methods and tools to address the problem of predicting future hospitalization periods (in days) for patients from a given set of historical patient data. The data mining techniques that we explored were linear regression, random forest and gradient boosting. For each technique, we used different historical data sets. The combination of data mining techniques and historical datasets enabled us to compare access and choose the combination which provides the best prediction of hospitalization period of a set of patients. Based on the results of our work, the random forest technique provides the best prediction of patient hospitalization.