Application of K-Means Algorithm on Normalized and Standardized Data for Type 2 Diabetes Subclusters

Current discovery indicates that Type 2 Diabetes (T2D) could be categorized into many sub-clusters, which is a step towards precision medicine. Implementation of feature scaling to cluster T2D into subgroups is crucial, aiming to transform and fit the data within a specific scale. This paper aims to compare the differences between cleaned, normalized, and standardized data in k-means clustering using T2D patient data. Two data transformation approaches were applied on the clustering algorithm, namely data normalization and data standardization. By comparing the clustering analysis results, normalized data (Elbow method (EM) =435.63) illustrates the best data point distribution to form clusters and good internal clustering validation scores in comparison to cleaned data (EM =502,254.97) and standardized data (EM = 23,518.82). We concluded that data normalization in the k-means clustering algorithm is the best method for data transformation for T2D sub-clustering compared to cleaned data and standardized data.