Feature weighing for efficient clustering

In cluster analysis, current algorithms assume that all features in the data contribute uniformly in assigning samples to clusters. This assumption can lead to poor clustering results, due to the existence of noisy and less important features. Feature weighting overcomes this issue by assigning different weights to features based on some notion of importance. According to feature weighting, more important features are assigned higher weights and less important features lower weight. When weights are applied to the data, the normal feature space becomes a transformed feature space. This paper proposes that the degree of transformation is as important as the weights themselves to find better clustering outcomes. A fixed feature weighting technique is used where the magnitude of weights is increased or decreased based on Cauchy sequence. The proposed methodology is tested on simulated and real world datasets using k-means and hierarchical clustering algorithms, and results are compared against clustering without weights.