Trajectory Mining for Laboratory Examination Data Sets

• Activity Number: 615
• Type: Topic Contributed
• Date/Time: Thursday, August 5, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics and Marketing
• Author(s): Shoji Hirano* and Shusaku Tsumoto+
• Companies: Shimane University and Shimane University
• Keywords: trajectory mining ; longitudinal data ; multi-source datasets

Abstract

This paper shows application of trajectory mining techniques to laboratory examination data, where dynamics of several laboratory examinations are represented as a trajectory in a multi-dimensional plane. Each trajectory is segmented and compared with other trajectories by multiscale matching, and clustered by using similarities obtained by matching technique. Comparison of trajectories grater than three-dimension is still a challenging problem because the zero-crossings of curvature (inflection points), that plays an important role in recognizing hierarchical structure of a trajectory, are difficult to be determined because curvature on a space curve is essentially sign-less. We propose to use the maxima of curvature instead of its zero-crossings for segmenting a trajectory, and demonstrate that our method could successfully capture the structural similarity of 3-D trajectories.