Use PCA to reduce dimensions before training a supervised classifier. Use clustering to create a "segment" feature that improves supervised model performance.