Unusual Sub-sequence Identifications in Time Series with Periodicity

Fast and intelligent data mining has recently become an integral part of data analysis and a pre-requisite for modeling. This is largely due to the introduction of more sophisticated data collection tools and the possibility of observing large datasets at increased higher frequencies. This paper aims to investigate the current methodologies used for the detection of time series discord sub-sequences and especially those with periodicity. A strategy will be suggested to use classical data mining techniques and statistical decision making to take advantage of the special features of the time series to make the detection more efficient and more objective. An entropy-based measure will also be introduced as an alternative to the Euclidean distance measure for identifying discord sub-sequences.