A Data Analysis Model Using Variable Time-Series Data for Health Care

P.Saravanakumar, S.Lina, N.Venkatesan

Abstract:  massive amount of time stamped data are generated in various domains such as health care, engineering, intrusion detection, stock market analysis and many more. This increasing availability, demands the analysis of temporal data for the purpose of prediction and classification. Time series classification is a challenging task due to the presence of high dimensional heterogeneous data; hampered with missing values and sampled at irregular time intervals. The project thus focuses on designing a classification model for multivariate time serious data that could be used for decision support system. In order to preserve the temporal characteristics of the data, the proposed methodology includes three phase. First, the missing values are imputed using Particle Swarm Optimization (PSO) based Inverse Distance Weighted Interpolation method (IDW). Secondly, the temporal data are transformed into high level symbolic representation (states and trends), through temporal abstraction based on Piecewise aggregation using dynamic window sizing. Finally, the abstracted states and trends are used as features to induce the Support Vector Machine (SVM) classifiers. The work is demonstrated using real dataset containing clinical trials of thrombosis patients and the results shows that the SVM classifiers has outperformed other classifiers with 95% classification accuracy.