Automated Requirement Dependency Analysis for Complex Technical Systems

Abstract

Requirements changes are a leading cause for project failures. Due to propagation effects, change management requires dependency analysis. Existing approaches have shortcomings regarding ability to process large requirement sets, availability of required data, differentiation of propagation behavior and consideration of higher order dependencies. This paper introduces a new method for advanced requirement dependency analysis based on machine learning. Evaluation proves applicability and high performance by means of a case example, 4 development projects and 3 workshops with industry experts.

Proactive Management of Requirement Changes in the Development of Complex Technical Systems

“…The validity of calculation was tested ex-perimentally and confirmed for requirement sets with 150, 250, 500 and 1,500 requirements. The computation time for each pair of requirements was determined for two representative computing capacities: 16 GB RAM/i7 CPU (regular laptop; Lenovo ThinkPad) and 13 GB RAM/P100 GPU (high-performance cloud server; Google Colab) [75].…”

“…Due to the limited extent of training data, dependency types are not distinguished by type in order to reduce the number of classes to be differentiated. Since the dataset contains a major class (None) with many entries and a minor class with few entries (Dependent), the criteria needed to be viewed macro averaged (equal weight on all classes) to be significant [75].…”

“…Table 4. Performance of dependency analysis for the articulated-arm robot case study [75]. The application for industrial case study on engine control unit revealed the importance of a context-specific dependency model.…”

“…In consequence, the importance of contextspecific data to train the dependency model is highlighted. However, the ability to reach high accuracy of dependency analysis for different contexts is also shown [75].…”

Requirement Dependency Extraction Based on Improved Stacking Ensemble Machine Learning

“…Machine learning has also been widely applied in cost prediction, software testing and software quality assessment in the software development process, such as in consistency research between developers and tasks [21], integration testing [22], software development cost prediction [23] and software quality assessment [24]. Meanwhile, requirements engineering has also applied a large number of machine learning methods [25][26][27][28][29][30][31][32][33][34][35][36][37][38][39], such as requirement acquisition, requirement formalization, requirement classification, the identification of software vulnerabilities from requirement specifications, requirement prioritization, requirement dependency extraction and requirement management. Previous studies have demonstrated that the automatic extraction of requirement dependency relationships is a feasible and effective task [32][33][34][35][36][37][38].…”