Predicting Delays in Software Deliveries using Networked Classification at ING

Abstract

Delays in the delivery of software projects and the corresponding cost and schedule overruns have been common problems in the software industry for years. A challenge within software project management is to make accurate effort estimations during planning. Software projects are complex networks, with multiple dependencies between software tasks.

This study aims to combine the field of effort estimation and networked classification to utilise network information for delay prediction in industry. We conducted a case study at ING, resulting in a number of insights with regards to networked classification in an industry setting.

There is a difference in the organisational structure of open-source and industry projects. This constitutes to a difference in available information, but there is also an opportunity to leverage the organisational structure of ING to improve delay prediction performance.

Using weights in networked classification has shown no improvement compared to not using them, but relational models do benefit from larger datasets as the used network contains more relational information.

Based on the insights we recommend ING to: keep track of more information, improve data quality by educating their teams and create models for specific domains or teams to leverage their organisational structure.