A framework for parameter free motif discovery process for improving the performance of time series signal classification

Abstract

Time series data have been generated virtually in a large volume in newlineevery field and they are stored in time series databases. Knowledge discovery newlinefrom data uses primitives such as query by content, classification, prediction, newlinemotif discovery and outlier analysis to analyze the importance of time series newlinedata. Motif discovery provides useful insights and hidden semantic to the newlinedomain experts and summarizes the importance of time series databases. They newlinehave been widely used in finance, healthcare and education. Despite there are newlinemany state-of-the-art discovery algorithms for motif discovery, they do not newlinetypically scale up and they are mostly user/task/parameter dependent. In newlineparticular to the application of cardiovascular disease diagnosis, the state-of-theart newlinediscovery techniques have many limitations such as usage of minimal data newlineand they are exclusively parameter dependent. This present research work newlineproposes the concept of maximal motif discovery process by integrating TRIE newlineand anti-monotone property to extract the motifs from the approximated time newlineseries sequence using a user-defined parameter minimum support threshold. The newlineproposed technique reduces the redundant candidate motifs through the newlineextraction of maximal motif and it improves the classification performance. The newlineapproximation of raw time series signals is exploited by the usage of Symbolic newlineAggregate approximation (SAX) technique from the state-of-the-art time series newlineapproximation technique. newlineDespite the proposal of maximal motif, to reduce the search space and newlinegeneration of candidate motif further, this research work proposes genetic newlinealgorithm based motif discovery process to extract motif from the approximated newlinesequences. This process utilizes anti-monotone property as an objective function newlinewith minimum support threshold of user-defined parameter. This algorithm newlineproposes addition as a new operator rather cross-over in traditional genetic newlinealgorithm to attain the global optimum at the earliest. newline