Optimized Data Clustering Technique for Categorical Data

In this book, a new clustering technique for categorical-data is introduced. Essentially, the effectiveness of a clustering technique is significantly determined by two aspects, the searching method and the proximity criteria. The proposed algorithm uses a genetic algorithm for clustering that is shown in the experiments to be an efficient clustering method for categorical-data. The proximity criteria adopt a rule-based information theoretical measure called weight of evidence. It finds the interesting patterns and measures the weight of these patterns that supporting the presence of an objective-value pair to be relevant to a cluster label. By summing up the total weight that the records acquire in the patterns due to presence of both the objective-value and the corresponding cluster label, the fitness in the chromosome is measured and hence how best the records are clustered together is seen.