Software engineers often construct quality-estimation models, used to predict the fault-proneness of software modules, by training a classifier from labeled software metrics data. They often encounter two challenges: noisy data and a lack of fault-proneness labels in real-world projects. You can?t train a classifier without fault-proneness labels. The clustering exploratory analysis method addresses these two challenges and uses clustering algorithms with the help of a software engineering expert. This method is unsupervised because it doesn?t require labeled training data to predict software modules? fault-proneness. Two real-world case studies verify this clustering- and expert-based approach?s effectiveness in predicting both software modules? fault-proneness and potentially noisy modules.