Similarity measure for heterogeneous multivariate time-series

Defining the similarity of objects is crucial in any data analysis and decision-making process. For those which effectively deal with moving objects, the main issue becomes the comparison of trajectories, also referred to as time-series. Moreover, complex applications may require an object to be a multidimensional vector of heterogeneous parameters. In that paper, we propose a similarity measure for heterogeneous multivariate time-series using a non-metric distance based on the Longest Common Subsequence (LCSS). The proposed definition allows for imprecise matches, outliers, stretching and global translating of the sequences in time. We demonstrate the relevance of our approach in the context of identifying similar behaviors of a person at home.