Line Centered Fuzzy Clustering for Time Series Prediction

Time series analysis and prediction has many useful applications in real life problems. Even, in predicting some natural events like solar activity and space weather its application is very essential. Moreover, it has many applications in command & control domain. Some examples in this area can be seen in (wireless) sensor networks, security and environmental monitoring. e.g., forecasting natural events which can help significantly to alleviate the causes of them and also it can help improve the effectiveness of disaster management process. Consequently, different approaches to solve the problem of time series prediction have been suggested.Fuzzy logic, neural network and neuro-fuzzy hybrid methods are among the most frequently used algorithms forthis purpose. In this paper, we propose an algorithm for predicting time series using fuzzy clustering. In this approach clusters are presented by lines as their centers. The suggested method has been tested with both artificial and natural datasets (Mackey-Glass and Natural Kp Series). In comparison with other researched methods we could register notable improvements in NMSE measure. Furthermore, the peak points of Kp are predicted with a higher ratio.