An Efficient and Accurate Time Series Classification Using Shapelets

Abstract

Time series data are sequences of values measured 
over time. One of the most recent approaches to classification of 
time series data is to find shapelets within a data set. Time series 
shapelets are time series subsequences which represent a class. 
In order to compare two time series sequences using shapelets, 
existing work uses Euclidean distance measure. But Euclidean 
distance has following limitation : it requires data to be 
standardized if scales differ. In this paper, we perform 
classification of time series data using time series shapelets and 
used Mahalanobis distance measure. And also we have 
performed pessimistic pruning on decision tree. The 
Mahalanobis distance improves the accuracy of algorithm and 
pessimistic pruning method reduces the time complexity of 
testing and classification of unseen data.  The Mahalanobis 
distance measure differs from Euclidean distance in that it 
takes into account the correlations of the data set and 
is scale-invariant. We show that our algorithm is much more 
accurate and faster than existing algorithms.