Comparison and Evaluation of Clustering Algorithms

This paper compared the effects of different clustering models and found some ways to evaluate the cluster models. Clearly, the Gaussian mixture model (GMM) and hierarchical clustering model (HAC) were studied. To determine the number of clusters, the Silhouette Coefficients line graph was plotted in model GMM, where the best number of cluster groups was found to be 4. However, the clustering effect is poor, as the scatter diagram shows. Next, a hierarchical clustering model was built, which has a better cluster effect. It is difficult to determine the clustering number in HAC, so the number of clusters in GMM is applied to HAC. It can be seen from the scatter plot that the clustering effect for model HAC is better. In summary, this essay combined the advantages of GMM and HAC, which makes it easier to determine the number of clusters. In addition, this essay used a way to evaluate the clustering models: draw the clusters after dimensionality reduction (PCA) in a 2D scatter plot.