The K-means algorithm is one of the most widely used unsupervised learning methods for clustering data. This research is inspired by improving its performance using a Metaheuristic based on the collective work of ants. K-means is easy to use due to its simplicity and ease of application. However, it tends to present weaknesses in certain information domains, especially in the overlapping and correct assignment of groups, generating poor quality clusters. A strategy focused on improving the quality of the clusters and minimizing the computational cost is proposed through an initialization method, performing experiments on a given data set.