Information Technology Reference
In-Depth Information
Fig. 1. Illustration of our framework
by representative subsequences. All the time series data instances are converted
to an
k by the learned codebook with k codewords. That is, we summarize
how many codewords are used for a time series data instance and conclude the
representation with the histogram of the codewords. Then we apply classifica-
tion methods, such as support vector machine, to the time series data with BoW
representation.
R
2 Bag-of-Words Model for Time Series Classification
Suppose a time series instance is originally represented by x i =( x 1 ,x 2 , ..., x p )
wherewehave p timestamps for each instance. Each time series instance x i is
associated with a class label y i for i =1 , 2 , ..., n and y i
where n is
the number of instances and C is the number of class labels. As we mentioned,
directly representing a time series in this p -dimensional space might be too spe-
cific to describe a time series appropriately. In this section, we will address how
to apply BoW for representation of time series data, and present the details for
the codebook learning and representation encoding.
∈{
1 , 2 , ..., C
}
2.1 Codebook Learning
Due to the intra-class variability, using global properties for times series data
might not be suitable for the classification. The local patterns could be applied
for improving the recognition performance. Therefore, we represent each time
series data instance by feature vectors derived from subsequences. It is worth
Search WWH ::




Custom Search