Feature Selection with Invasive Weed Optimization Clustering Algorithm

Dimensionality reduction is an important preprocessing technique in clustering domain. Feature selection is one of dimensionality reduction methods, in which it selects a subset of the most relevant features. This paper proposes a feature selection method based on Invasive Weed Optimization (IWO) algorithm. The IWO uses clustering strategy on data features (not data points). Also, the number of reduced features as an initial parameter for IWO is specified. This parameter is automatically determined using a method inspired by Principal Component Analysis (PCA). To see the effect of proposed “IWO feature selection clustering” (IFSC), both PCA and the IFSC outputs are applied to the K-means algorithm, separately. The proposed algorithm is tested on three different sizes of datasets from UCI machine learning repository. The obtained results show that the proposed IFSC outperforms significantly better than the PCA strategy.

Keywords:

Swarm Intelligence , Invasive Weed Optimization , Data Reduction , Data Clustering , Variation Size of data.