Final Capstone Topic: Predicting Incident Management Service Level Agreement (SLA) Failures

A Logistic Regression Application

Abstract

Information Technology (IT) Service Management practices optimize the efficiency and effectiveness of IT services delivered to users. Incidents represent service disruptions. Service Level Agreements (SLA) establish thresholds for resolution of incidents within specified timeframes based on impact and urgency designations. Decreasing SLA breaches increases the availability of IT services and represents an important consideration for IT service providers.

This study explores indicators of incident SLA breaches with respect to data available during the early stages of an IT incident’s lifecycle. The study built a Logistic Regression model using Python and a number of tools from the SciKit-Learn library. Some supplementary analysis leveraged the R language.

• Task 2 Data Analytics Report describes the data collection, the data extraction and preparation, and the analysis steps performed throughout the study followed by a summary of findings, implications, and recommendations
• Task 3 Executive Summary provides an overview of the study, results, and recommendations
• Task 3 Presentation serves as a summary of the study for discussion purposes

Data Source

A study investigating factors contributing to incident management SLA risk requires an extract from an ITSM system used by an IT organization for tracking incidents over a specific period. This study leveraged an existing, publicly-available data set used in the 2014 Business Processing Intelligence Challenge (BPIC) (see https://www.win.tue.nl/bpi/doku.php?id=2014:challenge). While the challenge released four data sets, this project focused only on the Incident Records file. The terms for use of the data set specify that “The user is allowed to remix, transform or build upon the data, but only for noncommercial purposes”.

Acknowledgements

Throughout the notebooks, links point to some of the resources used throughout development of the regression model. In particular, I found the information provided by Kevin Markham at https://www.dataschool.io extremely valuable.