Mining Temporal Patterns from Health Care Data

This paper describes temporal data mining techniques for extracting information from temporal health records consisting of a time series of elderly diabetic patients’ tests. Diabetes is one of the most common diseases affecting quality of life and is potentially life threatening for elderly people in developed countries. We propose a data mining procedure to analyse these time sequences in three steps to identify patterns from any longitudinal data set. The first step is a structural-based search using wavelets to find pattern structures. The second step employs a value-based search over the discovered patterns using the statistical distribution of data values. The third step combines the results from the first two steps to form a hybrid model. The hybrid model has the expressive power of both wavelet analysis and the statistical distribution of the values. Global patterns are therefore identified.