skip to main content

SciTech ConnectSciTech Connect

Title: Statistical model selection for better prediction and discovering science mechanisms that affect reliability

Understanding the impact of production, environmental exposure and age characteristics on the reliability of a population is frequently based on underlying science and empirical assessment. When there is incomplete science to prescribe which inputs should be included in a model of reliability to predict future trends, statistical model/variable selection techniques can be leveraged on a stockpile or population of units to improve reliability predictions as well as suggest new mechanisms affecting reliability to explore. We describe a five-step process for exploring relationships between available summaries of age, usage and environmental exposure and reliability. The process involves first identifying potential candidate inputs, then second organizing data for the analysis. Third, a variety of models with different combinations of the inputs are estimated, and fourth, flexible metrics are used to compare them. As a result, plots of the predicted relationships are examined to distill leading model contenders into a prioritized list for subject matter experts to understand and compare. The complexity of the model, quality of prediction and cost of future data collection are all factors to be considered by the subject matter experts when selecting a final model.
 [1] ;  [1] ;  [2]
  1. Los Alamos National Lab. (LANL), Los Alamos, NM (United States). Statistical Sciences Group
  2. ARDEC, Picatinny Arsenal, Township, NJ (United States)
Publication Date:
OSTI Identifier:
Report Number(s):
Journal ID: ISSN 2079-8954; PII: systems3030109
Grant/Contract Number:
Accepted Manuscript
Journal Name:
Additional Journal Information:
Journal Volume: 3; Journal Issue: 3; Journal ID: ISSN 2079-8954
Research Org:
Los Alamos National Laboratory (LANL), Los Alamos, NM (United States)
Sponsoring Org:
Country of Publication:
United States
97 MATHEMATICS AND COMPUTING automated model evaluation; variable selection, environmental exposure, system usage, advancing underlying theory