Building and Testing a Fuzzy Linguistic Assessment Framework for Defect Prediction in ASD Environment Using Process-Based Software Metrics

Abstract

The objective of the present work is to build and test a framework which makes use of process-based software metrics to determine the defects in software projects in an agile software development environment. A methodological framework based on fuzzy linguistic modelling has been proposed to predict the defect density using various process metrics derived from literature studies to measure the attributes stated in the agile manifesto. Further, the model is investigated by using the data set from the PROMISE software engineering repository, and its performance has been compared with existing models from the literature. The proposed model shows better accuracy (for projects with size ≥ 50 KLOC) as observed from statistical results, i.e. RMSE (18.69), NRMSE (0.0110), MMRE (0.0539) and BMMRE (0.0585). The value of R2 for all projects size up to 10 KLOC is 0.993, projects with size 10–50 KLOC is 0.998, and projects with size ≥ 50 KLOC is 0.997. The main contribution of the framework lies in the use of the process metrics and their linguistic assessment. Results obtained from the linguistic model emphasise the value of concepts related to customer involvement and interactions, the collaboration between stakeholders, responding to change, i.e. flexibility, team experience, skills, communication and coordination, as per agile manifesto.