Arbitrary length multivariate sequence similarity : a case study on north indian tropical cyclones

Abstract

In this project, I study and investigate how one compares arbitrary length multivariate data sequences by projecting the data sequences into a fixed low-dimensional space. To enable the comparison, a similarity value between two data sequences are computed using the Longest Common Subsequence (LCSS) algorithm for all possible pairs of data sequences, followed by the projection of the data sequences into a low-dimensional space using the ISOMAP algorithm. The contributions of my project is (i) an approach to choose the LCSS parameters to enable a good dimensionality reduction (i.e. similar data sequences are closed to one another, and vice versa), and (ii) application of the comparison approach to the 29 North Indian Tropical cyclones occurring from 2007 to 2011.