Application of Statistical Analysis Techniques to Computer Performance Evaluation.
Abstract
Multivariate statistical analysis techniques were examined to determine their applicability to computer performance evaluation. Specific techniques were multilinear regression, automatic interaction detection (AID), and ridge regression. A simulated data base created through a computer performance evaluation simulation developed at the Air force Institute of Technology was modeled using each technique. Multilinear regression derived the most accurate modeling equations, though no method adequately explained the variation in turnaround time. Ridge regression, designed to circumvent multicollinearity, also created modeling equations, but was less accurate because no multicollinearity was present in the system. AID, however, clustered the data into groups which explained high levels of the variability of the criterion variable within each group, and little between the groups. Data records which met the selection criteria for each group should also have values of the criterion variable within the range specified by that group. Although difficult to use, AID provided initial information about systems which could not be modeled through regression or other techniques. The applicability of each technique to selected types of computer systems was alsodocumented. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 1981
- Accession Number
- ADA115550
Entities
People
- Margaret A. Stover
Organizations
- Air Force Institute of Technology