Network system optimization using machine learning analysis: A systematic review

The contemporary digital infrastructure is based on network systems, which serve as the foundation of communication, data transfer, and service provision in various fields of application. As the network traffic, heterogeneous architectures, and quality-of-service (QoS) requirements have grown exponentially, the conventional methods of optimization have become progressively ineffective because of their lack of dynamism and computational complexity. Machine Learning (ML) has become a disruptive method of optimization of network systems as it allows making decisions that are data-driven, flexible, and intelligent. The current paper includes a systematic review of the newest developments in streamlining network systems with the help of machine learning. The review takes a systematic approach that implies identification, screening and analysis of peer-reviewed journal and conference articles published between 2021 and 2025. Supervised learning, unsupervised learning, reinforcement learning, and deep learning are considered in connection with network performance optimization aims like traffic management, routing, congestion control, resource allocation, fault prediction and energy efficiency. The findings suggest that the recent research is now dominated by reinforcement learning and deep learning models because of their ability to manage complex, dynamic, and large network environments. Nonetheless, the limitations of the current models, like data scarcity, model interpretability, scalability, and real-time deployment, are still major obstacles to their practical application. The article finishes with the future directions of research, systems such as explainable AI, federated learning, and hybrid optimization structures to increase the strength and applicability of machine learning-based network optimization systems.