Symbol Encoding Method of Time Series Data and Markov Model
Takafumi Katayama*, Eiji Suzuki* and Masao Saito**
* Dept. of Electronic Engineering, Kogakuin Univ.
** Institute of Medical Electronics, Univ. of Tokyo
A symbol encoding method is proposed for the purpose of data classification. The parameters of a sample are obtained by linear prediction analysis. Log likelihood ratios are computed between the parameters of a sample and several templates. A sample is classified to the template distance, and then the parameters of the template is modified. A symbol (e.g. an alphabetic character) is assigned to each template. The whole time series data are encoded into the stream of symbols. A Markov model which produced the same symbol series is constructed. A model parameter is represented as a structure of the time series data.
This article is published under a Creative Commons Attribution-NoDerivatives 4.0 Internationa License.