Linguistic Expression of Time Series and Its Applications to Retrieval and Machine Learning

我们有很多种时间序列的数据，如每日温度和股票价格。我们很可能通过语言表达来理解它们的全球趋势和局部特征。在这项研究中，我们制定了一种方法来提取语言表达的全球趋势，局部特征和振荡使用模糊集，并实现了一个系统，从时间序列数据中提取它们。然后，我们实现了两个原型系统，用于在时间序列数据库中检索相似的时间序列并从包括时间序列的一组数据中学习知识。相似时间序列的检索是基于全局趋势、局部特征和振荡的语言表达的相似度。知识的学习是模糊C4.5的扩展，包括全局趋势，局部特征和振荡属性。

We have many kinds of data of time series such as daily temperatures and stock prices. It is likely that we understand them via linguistic expressions for the global trend and local features. In this research, we formulate a method to extract linguistic expressions for the global trend, local features and oscillation using fuzzy sets and implement a system for extracting them from a time series data. Then we implement two prototype systems for retrieving similar time series in a database of time series and learning knowledge from a set of data including time series. The retrieval of similar time series is based on a similarity degrees of linguistic expressions for global trend, local features and oscillation. The learning of knowledge is an extension of fuzzy C4.5 to including attributes of global trend, local features and oscillation.