Measuring Cost Avoidance in the Face of Messy Data

摘要

This paper presents alternative methods to forecast or predict failure trends when the data violates the assumptions associated with least squares linear regression. Simulations based on actual case studies validated that least squares linear regression may provide a biased model in the presence of messy data. Non-parametric regression methods provide robust forecasting models less sensitive to non-constant variability, outliers, and small data sets.