Time series segmentation of linear stochastic processes for anomaly detection problem using supervised methods

The problem of time series segmentation for real-world applications has received much attention recently. Different industrial machines as elements of critical infrastructure for energy extraction, pumping, generation and other operations are equipped by hundreds of sensors which measure and evaluate variety data sets such as temperature, vibration, accelerations, pressure, voltage and so on. In many cases these measurements are unreliable, incomplete, inconsistent, and noisy and hence they can be interpreted as realizations of some linear stochastic processes. The task of recognizing of anomalous operation mode of machines can be reduced to the problem of pattern recognition, change point detection or segmentation in time series. In this paper we propose a general approach for time series segmentation of linear stochastic processes based on supervised learning algorithms which are machine learning algorithms using a mapping from input samples to a target attribute of the data. We also perform empirical examples for some hypothetical time series segmentation.