A hybrid type-2 fuzzy clustering technique for input data preprocessing of classification algorithms

Recently, clustering has been used for preprocessing datasets before applying classification algorithms in order to enhance classification efficiency. A strong clustered dataset as input to classification algorithms can significantly improve computation time. This can be particularly useful in Big Data where computation time is equally or more important than accuracy. However, there is a trade-off between speed and accuracy among clustering algorithms. Specifically, general type-2 fuzzy c-means (GT2 FCM) is considered to be a highly accurate clustering approach, but it is computationally intensive. To improve its computation time we propose here a hybrid clustering algorithm called KGT2FCM that combines GT2 FCM with a fast k-means algorithm for input data preprocessing of classification algorithms. The proposed algorithm shows improved computation time when compared with GT2 FCM as well as KFGT2FCM on five benchmarks from UCI library.