Understanding the Challenges and Needs of Requirements Engineering for Data-centric Systems

Sammanfattning: Background: As technology is advancing day by day, people tend to produce enormous volumes of data. This exceptional growth in data is leading to an increase in the development of intelligent systems that make use of this huge data available. We group the development of such type of intelligent software systems and term them as "Data-Centric Systems (DCS)". Such systems include AI/ML components in an aself-driving car, Recommender systems any many more. Developing DCS is complexin the Software development life cycle process; one of the main reasons behind this complexity is the ineffective handling of requirements. Moreover, the literature suggests that a large percentage (48%) of development problems begin during the requirements phase and fixing requirements-related problems consumes a high cost of rework in later stages of software development. To design DCS effectively, RE techniques are considered one of the essential factors, since it is required to promote the combination of a system’s functional and implementation expectations from two entirely different perspectives, such as customers and developers. Although RE frequently plays a critical role in DCS, little is known about how RE can effectively be incorporated into developing such systems. Objectives: This thesis aims to focus on understanding industry experiences in the development of DCS with the main emphasis on RE and investigate the techniques/approaches used in DCS designing during the RE process and identify the challenges practitioners face during the development process. Methods: Two workshop-style interviews are conducted to identify the design process of RE and the practitioners’ challenges during DCS development. To complement the results from the workshop and scaling up the target population, an online survey is conducted. Results: From the workshops, we have identified that no explicit stakeholder is involved during the RE phase of DCS. Still, all people collectively take the decisions when it comes to developing in agile, and the role varies depending on the type of projects the stakeholder is involved in. Four categories of requirements were identified, namely regulatory, infrastructure, data, safety and security requirements. Techniques/approaches used to elicit, document, analyse and validate the requirements were identified. Based on the data received, we have identified ten challenges faced by the practitioners during the DCS. Based on the number of responses recorded in the survey, the categorisation and the techniques/approaches used for RE were prioritised based on the highest number of responses received. A total of 15 themes were generated for the challenges based on the responses received from participants. Conclusions: To conclude, a specific RE architecture needs to be designed to help the practitioners during the development of DCS. We believe that the analysis of these insights provides the industry with a structured overview of the DCS development process in RE. Besides, this thesis allows the academic community to steer future research based on an understanding of industry needs in RE for DCS.