A Framework for an Automated Development Environment to Support the Data-driven Machine Learning Paradigm

• Published in: Proceedings of IEEE Southeastcon pp. 329 - 331
• Main Authors: Bowman, Anthony D., Prabhakar, Shyam P., Jololian, Leon
• Format: Conference Proceeding
• Language: English
• Published: IEEE 26.03.2022
• Subjects: development environment; Focusing; Machine learning; Machine learning algorithms; Natural language processing; Object detection; Parallel processing; Software; software development
• ISSN: 1558-058X
• DOI: 10.1109/SoutheastCon48659.2022.9764094

In recent years a machine learning paradigm has emerged, focusing on a data-driven approach as opposed to traditional development. Advances in machine learning techniques have allowed researchers to make substantial gains in tackling complex problems in diverse fields such as medical diagnosis through image analysis, object detection and tracking, and natural language processing. However, often researchers only employ one or two machine learning algorithms with a static feature set while only testing a single hypothesis. This self-imposed bottleneck often produces suboptimal results because it arises from using machine learning within the classical, algorithmic context using existing development tools. Therefore, there is a need to create new development tools which reflect this change to the machine learning paradigm. In this research, we propose a development environment that allows researchers to leverage those capabilities more fully by shifting not only the tool they use but also their mindset. Our proposed environment serves as an intermediate tool, guiding the researcher and making full adoption of the machine learning paradigm throughout the software development process easier. To accomplish this, our framework is defined by a three-layer structure designed for subject domain assessment, data manipulation and feature set exploration. Supported by parallelism, data cleaning and feature engineering, this research provides a conceptual basis for future creation of development environments for the machine learning paradigm. Future development of such a conceptual design would allow for additional intelligent tools to aid the user in designing solutions and support reusability at the design level.