This paper addresses the initialization issue in the k-means clustering algorithm, which can significantly affect its performance due to its sensitivity to initial cluster center placement. The authors propose a new approach based on discriminant analysis that improves two existing hierarchical initialization methods, var-part and pca-part, demonstrating that their technique competitively enhances clustering outcomes on a range of data sets. The findings indicate that the proposed modification not only maintains linear complexity but also improves the stability and performance of initialization in k-means clustering.