Indian Journal of Science and Technology

Abstract

Time series data are commonly used in data mining. Clustering is the most frequently used method for exploratory data analysis. In this paper a model is proposed for similarity search in recent biased time series databases based on different clustering methods. In recent biased analysis, data are much more interesting and useful for predicting future data than old ones. So in our method, we try to reduce data dimensionality by keeping more detail on recent data than older data. Due to “Dimensionality Curse” the original data is mapped into a feature space by means of Vari–segmented Discrete Wavelet Transform1 and then similarity measurement is performed by applying different clustering methods like Self Organizing Map (SOM), Hierarchical and K-means Clustering. This model is tested using Control Chart Data and the clustering result observed proves that the proposed model is better in grouping similar series under various resolutions.
Keywords: Clustering, Dimensionality Reduction, Discrete Wavelet Transform, Feature Extraction, Hierarchical Clustering, K-means Clustering, Self Organizing Map, Similarity Measurement