Date of Award

1-1-2015

Document Type

Dissertation

Degree Name

Ph.D.

Department

Computer Science

First Advisor

Anneliese Andrews

Keywords

Cluster Analysis, cost prediction, help desk, PCA, reliability, SRGM

Abstract

IT help desk operations are expensive. Costs associated with IT operations present challenges to profit goals. Help desk managers need a way to plan staffing levels so that labor costs are minimized while problems are resolved efficiently. An incident prediction method is needed for planning staffing levels. The potential value of a solution to this problem is important to an IT service provider since software failures are inevitable and their timing is difficult to predict. In this research, a cost model for help desk operations is developed. The cost model relates predicted incidents to labor costs using real help desk data. Incidents are predicted using software reliability growth models. Cluster analysis is used to group products with similar help desk incident characteristics. Principal Components Analysis is used to determine one product per cluster for the prediction of incidents for all members of the cluster. Incident prediction accuracy is demonstrated using cluster representatives, and is done so successfully for all clusters with accuracy comparable to making predictions for each product in the portfolio. Linear regression is used with cost data for the resolution of incidents to relate incident predictions to help desk labor costs. Following a series of four pilot studies, the cost model is validated by successfully demonstrating cost prediction accuracy for one month prediction intervals over a 22 month period.

Provenance

Recieved from ProQuest

Rights holder

Joseph David Lucente

File size

184 p.

File format

application/pdf

Language

en

Discipline

Engineering, Information science

Share

COinS