University of Twente Student Theses


A methodology to build interpretable machine learning models in organizations

Sajid, S. (2023) A methodology to build interpretable machine learning models in organizations.

This is the latest version of this item.

[img] PDF
Abstract:Data in an organization is table-stakes. But, knowing how to manage and effectively optimize the data to make better business decisions is what makes an organization set itself apart from the competition. One of the most compelling reasons for an organization to manage its data effectively and efficiently is to make better and informed business decisions. Data-driven insights provide organizations and stakeholders the required capabilities to generate real-time insights and predictions to optimize and drive better performance and decisions. These insights not only uncover opportunities to expand the organization’s growth in the industry, but also provides visibility into internal processes, core activities, critical business operations within the organization to make more concrete decisions. However, data in organization is far-reaching because it ranges from simple visualizations driven by analytics to in-depth insights of organizational performance. Moreover, with advances in technology and the many capabilities, there is a paradigm shift towards leveraging advanced analytics and Machine Learning (ML) related suite of techniques to make accurate and informative decisions. Despite the advantages and predictive insights that come with advanced analytic-based decision making, there is also a certain degree of accountability that needs to be maintained i.e. ensuring the decisions driven by analytics are interpretable to all stakeholders who are within the realm of decision-making. Among these stakeholders, those that are inclined towards business, more often than not, lack the technical knowledge to grasp the logic underpinning ML and analytical models. The solution may not necessarily be to simply find better ways to convey how an analytical system works; rather, it's about creating a transparent, standardized and interpretable suite of processes that can assist even a data-related stakeholder to build and understand the outcome at ease and then convey it to others. In order to effectively address these concerns, this study provides a comprehensive systematic literature review to understand the state-of-the-art and best practices to introduce interepretability for a business context. The results are then complemented with qualitative data collection processes such as interviews with relevant stakeholders to understand existing methodologies and the need to establish a standardized methodology to achieve the objective. As a result of the data collection process, the information has been consolidated to design and develop a business-friendly methodology that bridges knowledge gaps and the limitations identified to build better and interpretable ML models in a business setting. Further, the practicality of the proposed methodology is then demonstrated by the researcher in a real-world case study. The researcher has also validated the perceived usability with a series of experts in the domain to understand if the methodology would prove to be effective and efficient in forthcoming use cases.
Item Type:Essay (Master)
NXP Semiconductors, Eindhoven, Netherlands
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Programme:Business Information Technology MSc (60025)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page