A Hybrid Approach for Linguistic Summarization of Time Series

Linguistic summarization is an important step to extract information from a time series in an efficient and effective manner that simulates the human perspective. Before performing this process, scientists suggested using time series representations to identify the trends, then summarize the characteristics associated with these trends. In this paper, we show how to efficiently adapt and implement a piece-wise linear representation of time series to summarize the dynamic characteristics of trends. First, we show how to build time series representation using a modified Bottom up algorithm. Then we use a set of features to characterize the trends. Based on the protoforms proposed by Yager and the classical Zadeh's calculus of linguistically quantified propositions, we derive the linguistic summaries and their measures of quality. The experimentation done on real data show interesting and promising findings.