CENTAURUS: A Cloud Service for K-means Clustering

We present CENTAURUS, a scalable, easy to use, cloud service and pluggable framework for k-means clustering that automatically deploys and executes multiple k-means variants concurrently, and then scores them to provide a clustering recommendation. CENTAURUS scores clustering results using Bayesian Information Criterion to determine the best model fit across cluster results. CENTAURUS visualization and diagnostic tools help users interpret clustering results. We empirically evaluate CENTAURUS and compare it to MZA, a popular desktop tool that uses k-means clustering to extract farm management zones from soil electroconductivity data. We show that CENTAURUS produces better results, is more scalable, and requires less guidance from the user.