Developing Machine Learning Skills With No-Code Machine Learning Tools

Introduction

Building a machine learning (ML) model involves multiple complex skill sets. In the most minimal form, the standard protocol for building and using ML models involves at least three steps. Firstly, data must be collected and preprocessed. Secondly, a model must be trained using the preprocessed data, and thirdly, the trained model must be deployed in some form of application (Ramos et al., 2020). Generally, each step in this process requires advanced technical skills such as data collection, data preprocessing, model training, and model performance evaluation. In addition to their complexity, these required skills are cognitively demanding to acquire and use and are thus often restricted to experts. If we take into account the fact that, most models have to be updated regularly to accommodate for evolutions in the data (concept drift), the entire ML process then becomes expensive and unattainable for organizations and individuals without extensive resources and skills.

Against this backdrop, No-code ML (NML) tools have emerged as avenues for individuals to build ML models without possessing the requisite technical knowledge and skills (García-Ortiz and Sánchez-Viteri, 2021). This is because NML tools commonly exist as visual programming tools for ML (von Wangenheim et al., 2021). Hence NML tools considerably reduce the cognitive effort needed to create ML models. The reduction in effort is achieved because users focus on the logic of the system being developed instead of the textual elements (programming language syntax and semantics). More specifically, NML tools allow users to train ML models either by using a drag-and-drop interface to place visual elements on a canvas; or by specifying input, output parameters, and values in a few clicks. Furthermore, where visual elements are employed, they are largely used in the form of blocks or flows. This approach improves users' ability to learn by helping them prevent errors and improving their understanding of the concepts at hand. Consequently, NML tools allow one to build models relatively quickly. Additionally, NML tools reduce the financial barriers to using ML. This is mainly due to their features, mode of packaging, and distribution. Quite a substantial number of NML tools are free to download and free to use. Also, some are available over the web as cloud-based tools, and as such, they substantially reduce the necessity of using specialized hardware for training.

Naturally, the emergence of NML tools has generated discourse in academic research. Extant literature highlights the potential of no-code ML tools in developing relevant ML skills, knowledge, and attitudes. For instance, Lao (2020) provides empirical evidence of the use of NML tools in teaching high-school students how to create ML models and troubleshoot their performance. Similarly, García et al. (2020) and Rodríguez-García et al. (2021) identify ML knowledge and computational skills as possible learning outcomes of introducing NML tools to high school students. The foregoing indicates that ML skills can be learned using NML tools. Despite this invaluable knowledge, what remains unclear is how NML tools contribute to the acquisition of ML skills. From a technology affordances perspective, the authors conceive that there is a gap in the explanation of how the affordances generated when learners interact with NML tools, lead to the development of ML skills. An explanation that fills this gap exposes the positive mechanisms embedded in NML tools that lead to the generation of ML skills. Subsequently, developers of NML tools will be able to identify and grow these mechanisms to maximize learning outcomes. Thus, in this chapter, we draw on the technology affordance theory to explain how NML tools afford the development of ML skills.

This chapter has six sections. The first section provides the rationale and motivation for this chapter, while the second provides an overview of NML and reviews relevant literature on the subject. The third section expounds on the theory of technology affordances and constraints while the fourth section outlines our methodology for this work. The fifth, and sixth sections present the core findings of the chapter by covering the results, analysis, and discussions respectively. To conclude, we identify avenues for future research and summarize our arguments and findings.