Advisory Committee Chair
Leon Jololian
Advisory Committee Members
Arie Nakhmani
Maqbool Patel
Murat Tanik
Earl Wells
Document Type
Dissertation
Date of Award
2022
Degree Name by School
Doctor of Philosophy (PhD) School of Engineering
Abstract
istorically, research has often been conducted in a hypothesis-driven manner with software development methodologies created to support those efforts. However, in recent years data-driven approaches to research have seen a dramatic rise in prominence. While software development methodologies such as agile development, extreme programming, and the waterfall model have allowed developers to tackle increasingly complex problems, they were not designed to efficiently support data-driven approaches such as the machine learning paradigm. To address the need to support the different programmatic requirements of both classical, hypothesis-driven as well as data-driven development, novel development strategies are warranted. In this research, we adapted the well-established spiral model to support both hypothesis- and data-driven development within its iterative design. Within this model, we included a novel framework for embracing the machine learning paradigm. By removing artificial limitations in the number and selection of machine learning algorithms and feature sets we have often seen in previous literature, this framework allows for the expanded application of machine learning techniques. Supported by parallelism, feature engineering, and the reuse of data and feature subsets, this framework supports the efficient exploration of both the problem and solution spaces. To demonstrate its benefits, we applied this updated lifecycle model to a complex neurological problem. ii The results from this case study show this lifecycle model now provides greater flexibility for the developer in tailoring solutions to the ever-changing needs of a project, be they hypothesis- or data-driven. Our framework allows for greater adoption of the machine learning paradigm, providing support for developers to efficiently expand the scope of their work while generating more optimal results
Recommended Citation
Bowman, Anthony D., "A Hybrid Data and Hypothesis-Driven Model for Software Development in Support of the Machine-Learning Paradigm" (2022). All ETDs from UAB. 201.
https://digitalcommons.library.uab.edu/etd-collection/201
