A Review of Modern Database Systems for Data Science Applications
TL;DR
The study highlights that no single database solution fits all scenarios, and hybrid architectures with polyglot persistence are increasingly adopted, offering guidance for selecting appropriate systems in data-driven environments.
Abstract
Modern data mining applications require scalable and high-performance database systems capable of handling structured, semi-structured, and unstructured data. Traditional RDBMS, designed for transactional consistency, face limitations in scalability and performance for big data analytics. As a result, modern systems such as NoSQL, NewSQL, distributed file systems, and cloud-native platforms have emerged, offering features like horizontal scalability, schema flexibility, and real-time processing. This paper provides a comprehensive overview and comparison of these database systems, focusing on their architecture, data models, and suitability for analytical workloads. It also examines their role in data science pipelines, including data ingestion, preprocessing, model training, and deployment. Key trade-offs based on the CAP theorem are discussed, along with performance metrics such as scalability, latency, and fault tolerance. The study highlights that no single database solution fits all scenarios, and hybrid architectures with polyglot persistence are increasingly adopted. The paper concludes by identifying research gaps in database optimization for machine learning, data governance, and real-time analytics, offering guidance for selecting appropriate systems in data-driven environments.