A novel subspace clustering method based on data cohesion model

The clustering of categorical data faces a series of challenges: full space based, sensitive to the sequence of input data and susceptible to the input parameters, any one of them can greatly affect the result. For those problems, we put forward a novel subspace clustering method based on data cohesion model (SCDCM). Inspired by the law of universal gravitation, we consider the clustering problems from the physical point of view. We define the data cohesion force and provide a data cohesion model. As the experiment result shows, our work have a high effect both on synthetic data and real world data. Therefore, the SCDCM is suitable for categorical data.