Cluster Analysis in Data Mining

A very good course, it gives me a general idea of how clustering algorithm work.

Very informative lectures, wonderful assignments. This course isn't so easy but it gives you real knowledge and useful experience.

Very detailed introduction of Clustering techniques.

Good, thorough coverage -- for a 4-week course -- of how to cluster. I liked the evaluation of clustering topic especially. Very few other instructors seem to discuss the vitally important evaluation of clustering results in any depth when they teach clustering. Dr. Han explained a comprehensive framework for understanding the effectiveness of any clustering system. I had never seen some of this material before, even though clustering was a topic appearing in a couple of other data science or machine learning courses that I have taken in the past. Ideally I would even wish to see this course extended to 6 or 8 weeks, so that case studies on difficult real datasets can be clustered. For example I had a terribly difficult ordeal last year before I took this course, trying to cluster the Kaggle.com dataset of the BOSCH competition. It has about 90% missing data in every row, and there are 2 million rows in total, and about 4500 columns! Kaggle's BOSCH is a SUPER tough dataset to work with! I hope to come back to try the BOSCH dataset again using my new knowledge of clustering some time soon. The reason I chose to run unsupervised clustering on this BOSCH dataset, which is ostensibly intended for supervised learning, is to eliminate significant amounts of the missing data from being exposed to multiple individual supervised learning models by prior clever grouping of examples. I am still postulating to the current day that clustering and creating another unique supervised learning model for each cluster is the most important step to eliminating missing data in this particular problem.