Distributed AutoML of Incremental Machine Learning Algorithms

AutoML for incremental models necessitates continuous monitoring and model management due to the evolving nature of streaming data. This paper describes a comprehensive distributed framework for the AutoML of incremental machine learning algorithms, specifically designed for high volume structured data and time-series problems. The library enables rapid prototyping for training incremental models and generates dynamic AutoML pipelines to ensure consistent training and monitoring of these incremental learning processes. The performance of the proposed framework is demonstrated through its application to two distinct case studies: (1) wind power forecasting, where it facilitates accurate energy bid predictions; and (2) scalability evaluation, which leverages the NYC taxi dataset to assess the framework's capacity for handling large-scale data.