In fuzzy clustering, initialization of cluster centers and membership degree matrices is one of the mandatory tasks for starting the fuzzy partitioning algorithms. Quality initialization of these matrices provides fast convergence to the final cluster centers. This reduces the required number of iterations and consequently results with low computational costs. In this paper, a novel algorithm is proposed to generate initial membership degree matrix for starting fuzzy partitioning clustering algorithms such as Fuzzy C-means, Possibilistic Fuzzy C-means and other variants of these algorithms. The proposed initialization algorithm is based on the selection of the feature with the highest variability among the others in a multidimensional numeric dataset. The experimental results show that the proposed algorithm performs well in reducing the number of iterations needed to converge to the final cluster centers for the examined synthetic and real datasets.