There is an increasing interest in techniques for analysis of time-related data in the last decades. Such interest is explained by the fact that nearly all human activities produce time-related data. In this project, we aim at developing time series classification algorithms. Although a great deal of classification algorithms has been proposed in the last few decades, experimental evaluation suggests that a simple K-nearest neighbour classifier outperforms more complex classifiers in a wide range of application domains.One currently open question is how one can improve the performance of the K-nearest neighbours classifier for application domains in which this technique typically underperforms. The main hypothesis of this research work is that a simple yet effective change of data representation may provide a meaningful improvement on the classification performance. Our goal is to investigate new, effective time series data representations and propose distance measures that are effective for such representations.
News published in Agência FAPESP Newsletter about the scholarship: