Abstract
Time series classification is a supervised learning problem that aims at labelling time series according to their class belongingness. Time series can be of variable length. Many algorithms have been proposed, among which feature-based approaches play a key role, but not all of them are able to deal with time series of unequal lengths. In this paper, a new feature-based approach to time series classification is proposed. It is based on ARIMA models constructed for each time series to be classified. In particular, it uses ARIMA coefficients to form a classification model together with sampled time series data points. The proposed method was tested on a suite of benchmark data sets and obtained results are compared with those provided by the state-of-the-art approaches.