K-Means is a widely recognized classical algorithm used for solving clustering problems in data analysis and various other applications. The predefining of number of clusters and the choice of initial cluster centroid limits its application in solving automatic clustering problems. Optimum number of clusters and optimal clustering solutions are automatically determined using internal cluster validity indices in Evolutionary K-Means algorithms. However, most validity indices are designed to address certain data types, thereby making their performance data dependent. As a result, selecting a specific internal validity index may critically affect the quality of the resultant clusters. This study aims to evaluate the performance of different cluster validity indices on the Evolutionary K-means framework with reference to different datasets of varying structure and characteristics. Fifteen internal validity indices are evaluated on Enhanced Firefly-K-means as a representative evolutionary K-means for automatic clustering. This study provides valuable insights into selecting appropriate fitness functions for evolutionary K-means algorithms for automatic clustering tasks. The experimental result shows that, on average, Calinski-Harabasz (CH) and Silhouette indices performed better than others on both real-life and synthetic datasets.