Integrating type-2 fuzzy logic system with fuzzy C-means clustering for weather prediction

• Main Author: Soozaei, Ahmad Shahi
• Format: Thesis
• Language: English
• Published: 2011
• Online Access: http://psasir.upm.edu.my/id/eprint/27714/1/FSKTM%202011%2030R.pdf; http://psasir.upm.edu.my/id/eprint/27714/

Today’s world concerns more about the impact of weather in the development of society. An accurate weather prediction system can act in as a vital role for making crucial decision on the life and property issues. In this research, the temperature attributes of weather is considered. Accurate weather temperature prediction can be achieved with regards to the quality of data processed. Fundamentally, weather prediction is complex due to the heterogeneous and enormous data. There are also factors such as outliers, noise and overlapped data which cause an increase of uncertainty in the data. Therefore, the assurance of data quality is associated by isolating these uncertainties factors. The quality of data is foreseen to increase the accuracy of prediction. However, most researchers in this domain do not consider the importance of data quality in their researches. In the existing prediction methods, Type-2 fuzzy logic is the proper method to deal with the uncertainty. In fuzzy systems, the relation between uncertainty of input data and fuzziness is expressed by membership functions. However, if the regions of the data of different classes are highly overlapping or contain noise and outliers, the value of membership function will be misleading. This effect is known as the membership un-robustness. Furthermore, the result or decision produced will not be accurate and lead to false prediction. Thus, overlapped data and uncertainty are two important issues which affect the quality of data. In this thesis, a method is proposed to predict next temperature value with high accuracy. The proposed method is based on combination of statistic equation with Fuzzy C-Mean (FCM) clustering and Type-2 fuzzy logic system (Type-2 FLS) with gradient descent algorithm. The statistic equation with FCM can be applied to handle outliers and cluster desired data and gradient descent in Type-2 FLS is utilized to tune the membership function parameters. Another feature of the proposed method is improvement in the performance time (run time) by clustering the desired data. The proposed method has been validated by experiments using Italy and New York weather temperature dataset. The findings show that the accuracy of this method for prediction next value increased as compared to base method. The accuracy percentage of proposed method on the Italy dataset was found to increase accuracy up to 89.6%. For New York dataset, the proposed method was found to increase accuracy up to 91% as compared to 67% by the base method. The performance time of the proposed method has improved 52% and 49% in comparison to base method for Italy and New Yorkdataset respectively. The results prove that the proposed method is more efficient than base method in accuracy and performance time basis.