Trends summarization of times series: a multi-objective genetic algorithm-based model

Abstract

Aim: The explosion of the amounts of data generated in many application domains, makes the paradigm of data summarization more essential. Furthermore, it is of a great interest to effectively handle some specific needs. In this work, we discuss an advanced model to drive linguistic summarization in the context of time series. This model relies on a multi-objective genetic algorithm mechanism to generate a set of best summaries from a large number of candidates. Methods: To achieve this objective, the current work is divided into two parts: The first part is dedicated for extracting the linguistic summaries of the dynamic characteristics of the trends of time series. It is achieved using the traditional genetic algorithm where the fitness function represents the truth degree of the linguistic quantified proposition. The second part is devoted to formalise the problem of interest as a multi-criteria optimization problem. We use different quality measures of summary as targets for improving the predicted set of summaries. To reach this goal, we use the Fast Non-Dominated Sorting Genetic Algorithm NSGA-II. Results: We evaluate the proposed approach on real data from a Smart Campus application (Neocampus project of the University of Toulouse, France). The results are promising and confirming the usability of the proposed approach. Conclusion: The proposed approach overcomes the problem of the overabundance of irrelevant linguistic summaries of the time series. It allows selecting a set of best summaries regarding some relevant criteria.