Pattern recognition using enhanced non-linear time-series models for predicting dynamic real-time decision making environments

Abstract

The abundance of data and importance of knowledge extraction to foresee the future has made time dependent data analysis an inevitable and challenging task in all areas of science and engineering. High dimensionality and the presence of noise in the non-linear time-series data makes it difficult for the existing clustering algorithms to produce efficient results. Hence, two approaches for time series representation (TSR) techniques by name hybrid dimensionality reduction (HDR) and extended hybrid dimensionality reduction (EHDR) and high low non-overlapping (HLN) clustering algorithm that produces efficient results by controlling noise and reducing the dimensionality optimally are proposed. A comparison of the experimental results on intraday non-linear stock data sets to predict the similarity in their intraday behaviour using K-means clustering algorithm with MINDIST as distance measure using symbolic aggregate approximation (SAX) and HLN using HDR and EHDR has proved that EHDR and HDR TSRs outperforms the other models.