Big Data Management for Machine Learning from Big Data.

The world is dynamic, and so are big data. The evolving challenges of managing big data volume and velocity have resulted in several studies focusing on machine learning models. Despite the usefulness of these models, further explanation is often required to interpret, understand, and effectively use the outcome of machine learning models. In this paper, we examine challenges of machine learning models in processing big data. These include the inherent uncertainty in data collection and questionable validity of machine learning model outcome. Motivated by the challenges arising from complex varieties due to the rigid schema required by the prevalent relational database model and data warehouse, we present (a) an architectural design of a schema-less big data repository aiming at capturing all data type (e.g., structured, semi-structured, and unstructured data) and (b) a data-driven approach to metadata collection for managing the big data.