Human Activity Recognition (HAR) plays a pivotal role in diverse fields such as healthcare, performance monitoring, and risk prevention, employing Machine Learning (ML) and Deep Learning (DL) algorithms. This paper introduces B-HAR (Baseline-HAR), an open-source framework based on Service-Oriented Architecture to facilitate the engineering and evaluation of different ML/DL-based HAR methodologies. By automating the creation and implementation of a baseline workflow, B-HAR enables researchers to assess and compare HAR methods effectively. It integrates prevalent data-processing techniques and popular machine and deep learning models, ensuring consistency in data preprocessing while allowing for custom model integration. The framework's efficacy is demonstrated across nine prominent HAR datasets, encompassing various sensor types and placements, showcasing its utility in engineering applications, particularly in healthcare, where it aids in diagnosis, rehabilitation, and treatment optimization for neurological and physiatric disorders, as well as assisting individuals with special needs.
