Exploring User Adoption and Experience of AutoML Platforms: Learning Curves, Usability, and Design Considerations

Abstract

<title>Abstract</title> Human daily activities and businesses generate a high volume of data which are expected to be transformed for the benefit of businesses and mankind. Organizations make use of machine learning platforms to make informed decisions from well gleaned insights of their real-time data. The process of learning machine learning is seemingly not an easy one, making it tedious for employees to learn easily and quickly. Meanwhile, the introduction of automated machine learning (AutoML) has made this easier. However, it is essential to understand how users embrace and implement the AutoML platform for their real-world problems. To achieve this, we conducted a quantitative study with 38 users focusing on understanding firstly, the learning curve (i.e. the alignment of users’ performance proportionately with the time required to complete a given machine learning task at any given time) and experience of people in the process of learning machine learning. Secondly, the strengths and weaknesses in the design and usability of ML and AutoML. Thirdly, the gaps in the usage or user experience of a newbie - an inexperienced or fresh entrant in the machine learning domain- an inexperienced and fresh entrant in the machine learning domain and professionals, and fourthly, the design factors needed to improve the user experience. Our findings revealed that users have high expectations in the usability of AutoML. In this study, we were able to carry out an assessment of awareness rate of AutoML among the respondents, users’ learning curve in AutoML environment, usability assessment as it relates to variations in users of AutoML, human computer interactivity in relation to in terms of user centeredness and experience of the environment. The study revealed the level of awareness, reasons for apathy and some usability concerns begging for improvement to attract a high rate of usefulness and adoption. In the near future, we hope to take this work further by engaging frequent users of various AutoML environments to ascertain the level of satisfaction using such platforms and identify areas of concern.