Analysis of certain nonlinear time series systems using soft computing techniques

Abstract

Nonlinear time series prediction has been a challenging task and important area of research in all branches of science and technology Though several techniques are used for the pattern prediction problem identifying unknown valid information such as patterns and relationships from large time series databases is difficult due to the presence of noise high dimensionality and non stationarity The temporal behavior of the time series data makes it difficult for direct usage in the application Clustering related items together is used to find similarity in the behavior of time series data High dimensionality newlineand the presence of noise in the nonlinear time series data makes it difficult for the existing clustering algorithms to produce efficient results Therefore a time series representation that takes into account the internal structure of the time series data is essential Hence two time series representations by name Hybrid Dimensionality Reduction Extended Hybrid Dimensionality Reduction and High Low Non overlapping newlineclustering algorithm are proposed in the first work The proposed works produce efficient results by controlling noise and reducing the dimensionality optimally The experimentation was carried out on intraday nonlinear stock datasets to predict the similarity in their intraday behavior A comparison of the experimental results using K means clustering algorithm with Euclidean and minimum distance distance measures using Discrete Wavelet Transform and Symbolic Aggregate approXimation newlinerespectively and HLN using HDR and EHDR has proved that EHDR and HDR TSRs outperforms the other model newline newline