How do you handle outliers in a dataset that might represent actual mechanical failures at Halliburton?
