Assisted Requirements Selection by Clustering Using Analytical Hierarchical Process

Abstract

The success of a project depends on the efficient prioritisation of its software requirements. The application of clustering and related data mining techniques for requirements prioritisation within the context of software engineering is still unexplored and frequently overshadowed by established procedures. This study begins a thorough investigation of clustering's untapped potential as a cutting-edge method to enhance requirements prioritisation and enhance project outcomes. To improve the organisation of complicated requirements and determine their relative importance, the study offers the novel idea of combining clustering techniques with the Analytic Hierarchy Process (AHP). Two meticulously constructed quantitative datasets, each containing 20 and 100 software meticulously form the core of this research. Notably, the development of an AHP dataset represents a fresh contribution and serves as a standard by which clustering methods can be unbiasedly assessed. Five main clustering algorithms emerge as the investigation progresses: K-means, Hierarchical, Partition Around Medoids (PAM), Gaussian Mixture Models (GMM), and BIRCH. Each of these methods offers a wide range of analytical techniques for examining the datasets. The Dunn Index, Silhouette Index, and Calinski Harabaz Index are used to statistically measure the quality and cohesion of the created clusters to assess the effectiveness of these approaches. The MoSCoW approach is then used to order the identified criteria into clusters, guaranteeing that crucial requirements are met while allowing for flexibility for less important features. This dual strategy combines strategic prioritisation with quantitative analysis, allowing for an unbiased evaluation of clustering results and simplifying resource allocation based on requirement priority. Overall, this research pioneers the innovative integration of advanced data analysis methodologies into project management and emphasises the viability of clustering techniques for requirement prioritisation in the software domain, with a focus on the ground-breaking combination of AHP and clustering as a transformative approach to prioritise requirements.