An automatic methodology for the quality enhancement of requirements using genetic algorithms

Abstract

Context

The set of requirements for any project offers common ground where the client and the company agree on the most important features and limitations of the project. Having a set of requirements of the highest possible quality is of enormous importance; benefits include improving project quality, understanding client needs better, reducing costs, and predicting project schedules and results with greater accuracy.

Objective

This paper's primary goal is to create a methodology that can provide effective and efficient solutions for modifying poor requirements integrated into a full-fledged system, extracting the main features of each requirement, assessing their quality at an expert level, and, finally, enhancing the quality of the requirements.

Method

In the first step, a machine learning algorithm is implemented to classify requirements based on quality and identify those that are the likeliest to be problematic. In the second step, the genetic algorithm generated solutions to enhance the quality of the requirements identified as inferior.

Results

The results of the genetic algorithm are compared with the theoretically optimal solution. The paper demonstrates the significant flexibility of genetic algorithms, which create a wide variety of solutions and can adapt to any type of classifier. From the initial dataset of requirements, the genetic algorithm finds the optimal solution in 85% of cases after 10 iterations and achieves 59.8% success after only one iteration.

Conclusions

Genetic algorithms are promising tools for requirements engineering by delivering benefits such as saving costs, automating tasks, and providing more solid and efficient planning in any project through the generation of new solutions.

Introduction

At the beginning of any IT project, there is always a specifications phase that is typically companied by inconveniences and challenges. These arise because the client and service provider have very different views and knowledge levels about the project. The primary goal of requirements engineering is to solve these issues.

Requirements engineering helps the service provider obtain a better understanding of the problem, a greater awareness of the client's needs, and a deeper comprehension of how end users will employ the software [1]. Thus, high-quality requirements are a key pillar in planning any software project; they can help avoid consequences, such as unnecessary changes due to a lack of comprehension of the client's needs, the appearance of a lack of professionalism if the final product does not fulfil the client's expectations, and erroneous estimations of the project's scope and thus costs.

A new methodology to assess the quality of requirements uses decision trees [2], [3], [4]. It takes advantage of the internal rules of these decision trees to find the most optimal solution for changing poor requirements. It deserves special mention since the final results of those studies serve as the starting point for this investigation. The main difference involves this study's reliance on genetic algorithms for generating new solutions, which offers a wider variety of options and more adaptability to a project's needs at the modest cost of extra time for calculation.

A system that classifies non-functional requirements in software projects through semi-supervised machine learning has already been reported [5]. It is also possible to find other valid approaches, such as a framework based on a methodology that guides the user through the process of creating high-quality requirements [6] and the use of genetic algorithms with two levels of approximation adapted to the demands of requirements engineering [7].

The importance of requirements engineering has been demonstrated by establishing a relationship between requirement quality and project performance and outcome [8].

The present study offers a methodology based on machine learning techniques for improving requirements identified as poor The main target is to provide a set of effective solutions that can address the problematic requirements. These solutions will be shown to the end user to choose that is most appropriate for each case. This is accomplished by combining genetic algorithms with machine learning classification.