A performance comparison of using principal component analysis and ant clustering with fuzzy c-means and k-harmonic means

Several clustering researches focus on the idea for achieving the optimal initial set of clusters before performing further clustering. This may be accomplished by performing two-level clustering. However, such an idea may possibly not either significantly improve the accuracy rate or well alleviate local traps; contrarily it usually generates abundant runtime consumption. Thereby, one may turn to focus on the relieving the problems of high dimensional, noisy data and hidden outliers. Such difficulties usually occur in real-world environment; and can seriously spoil the computation of several of types of learning, including clustering. This paper proposes a performance comparison using feature reduction based method, principal component analysis (PCA) and ant clustering algorithm combining with two particular fuzzy clustering approaches, fuzzy c-means (FCM) and k-harmonic means (KHM). FCM and KHM are soft clustering algorithms that retain more information from the original data than those of crisp or hard. PCA is employed as preprocess of FCM and KHM for relieving the curse of high-dimensional, noisy data. Ant clustering algorithm is employed as the first level of clustering that supplies the optimal set of initial clusters to those soft clustering methods. Comparison tests among related methods, PCA-FCM, PCA-KHM, ANT-FCM and ANT-KHM are evaluated in terms of clustering objective function, adjusted rand index and time consumption. Seven well-known benchmark real-world data sets are employed in the experiments. Within the scope of this study, the superiority of using PCA for feature reduction over the two-level clustering, ANT-FCM and ANT-KHM is pointed out.