Clustering of clinical multivariate time-series utilizing recent advances in machine-learning

Abstract

The purpose of this thesis is to set the groundwork for future research on developing a machine-learning based anomaly detection system for hospitalized patients. Our first step was to study and analyze the project’s needs, background, and literature examining similar criteria. In the second step, we interviewed medical experts and researchers. Based on our research and the suggestions received in our interviews, we explored methods that could be utilized to approach the issue based on the data we collected. The results of these approaches were then discussed. According to the results, the K-means algorithm, which utilizes principle components to cluster, obtained the highest quality. We then discussed how other algorithms have been influenced more by the shape of the data than by the values of the data. Afterward, we made some suggestions about how this research could be approached in the future as we move forward.