Pattern Characterization in Multivariate Data Series using Fuzzy Logic - Applications to e-Health

Abstract

The application of classic models to represent and analyze time-series imposes strict restrictions to the data that do not usually fit well with real-case scenarios. This limitation is mainly due to the assumption that data are precise, not noisy. Therefore, classic models propose a preprocessing stage for noise removal and data conversion. However, there are real applications where this data preprocessing stage dramatically lowers the accuracy of the results, since these data being filtering out are of great relevance. In the case of the real problem we propose in this research, the diagnosis of cardiopulmonary pathologies by means of fitness tests, detailed fluctuations in the data (usually filtered out by preprocessing methods) are key components for characterizing a pathology. We plan to model time-series data from fitness tests in order to characterize more precise and complete patterns than those being currently used for the diagnosis of cardiopulmonary pathologies. We will develop similarity measures and clustering algorithms for the automatic identification of novel, refined, types of diagnoses; classification algorithms for the automatic assignment of a diagnosis to a given test result.