Fuzzy-based shapelets for mining climate change time series patterns

Abstract

It is difficult to identify visualized multi-climate change patterns from time series data due to the fact that the data begin to look similar over time. Traditionally, time series weather patterns are presented in the form of a linear graph, which is limited to discovering understandable climate change patterns. On the other hand, the Symbolic Aggregate Approximation (SAX) algorithm based on the Piecewise Aggregate Approximation (PAA), which is known as a popular method to solve this problem, has its limitations. Therefore, the aim of this research was to propose a fuzzy-based symbolic data representation, known as a Shapelet Patterns Algorithm (SPA), in order to come up with a Shapelet Pattern (SP) for climate change. The shapelet pattern was able to visualize climate change patterns in the form of coloured shapes to indicate annual changes in temperature patterns, such as cool, warm, hot and very hot. The experiment used the climate change data for 1985-2008 gathered from the Petaling Jaya station in the state of Selangor, Malaysia. The shapelet patterns revealed seven types of climate change patterns and presented detailed information on climate changes that can aid climate change experts in better decision making.