Human Input about Linguistic Summaries in Time Series Forecasting
Katarzyna Kaczmarek, Olgierd Hryniewicz, Rudolf Kruse · Advances in Computer-Human Interaction · 2015
Finding an appropriate predictive model for time series and formulating its assumptions may become very challenging task. We propose to represent time series in a human-consistent way using linguistic summaries. Such summaries describe general trends in time series and are easily interpretable for decision makers. The aim of this contribution is to show that the linguistic summaries may be successfully applied to support the analysis and forecasting of time series. Information about trends is first retrieved from experts, and then, processed with soft computing tools. The performance of the approach is verified on the real-world datasets from the M3-Competition. Users are asked to evaluate linguistic summaries that are intuitive and easy for interpretation. This paper shows that human-consistent summaries deliver new knowledge for forecasting. Keywords-information retrieval; human-computer interaction; time series and sequence models; Bayesian methods; supervised learning.