Eight years of AutoML: categorisation, review and trends

Abstract Knowledge extraction through machine learning techniques has been successfully applied in a large number of application domains. However, apart from the required technical knowledge and background domain, it usually involves time-consuming repetitive steps. Automated (AutoML) emerged 2014 as an attempt to mitigate these issues, making methods more practicable both data scientists domain experts. AutoML is broad area encompassing wide range approaches aimed at addressing diversity tasks over different phases discovery process being automated with specific techniques. To provide big picture whole area, we have conducted systematic literature review based on proposed taxonomy that permits categorising 447 primary studies selected search 31,048 papers. This performs extensive rigorous analysis field, scrutinising how addressed dimensions taxonomy, identifying any gaps remain unexplored well potential future trends. The yielded some intriguing findings. For instance, observed significant growth publications since 2018. Additionally, noteworthy algorithm selection problem gradually superseded by challenge workflow composition, which automates than one phase simultaneously. Of all AutoML, neural architecture particularly noticeable.