K-Means Clustering Algorithm for Mixed Data using Multidimensional Scaling

Traditional clustering techniques struggle when applied to mixed datasets that contain both continuous numerical and categorical features due to the presence of categorical data, which often leads to information loss and an increase in dimensionality. This study proposes a method that integrates the K-means algorithm with Multidimensional Scaling (MDS) to convert mixed datasets to continuous numerical representation by transforming categorical features into continuous numerical features, which results in a better clustering result. Pairwise distances were computed using Hamming distance for categorical features and Euclidean distance for numerical features before applying MDS. This approach addresses the limitations of the traditional way of handling categorical data using one-hot encoding. Experiments were conducted on varying real-world datasets, demonstrating the effectiveness of the proposed method in cluster separation and decreasing the sum of distances within the same cluster, resulting in a more accurate clustering tool for mixed data. This work contributes to the development of clustering methods for complex dataset analysis.