Reasoning support for predicting requirement change volatility using complex network metrics

Abstract

Due to inevitable design iterations, requirements are frequently changed and revised. If not effectively managed, undesired propagating changes may result in monetary and time losses, leading to project failure. Current requirement management tools lack formal reasoning for characterising change and its propagation. Based on prior research that suggests that requirement networks can be utilised to study change propagation and that requirement change volatility (RCV) can be measured through four classes: namely, multiplier, absorber, transmitter, and robust, which are defined based on how requirements behave to the initial change. This research investigates if complex network metrics can be used to predict RCV using computational methods. The RCV class metrics of each requirement are determined using industrial data and requirement relationships obtained from networks of a previously developed Refined Automated Requirement Change Propagation Prediction (R-ARCPP) tool. Complex network metrics are also computed for each requirement in the network. Regression analysis models, specifically ordinal output cumulative logit regression models employing information criterion based genetic algorithm model selection, and artificial neural network models, specifically backpropagation artificial neural networks, are employed to predict RCV using complex network metrics. Best performing models are discussed along with the limitations, conclusions, and future direction of this research.