A Zero-Touch Network Service Management Approach Using AI-Enabled CDR Analysis

Federated Learning based Anomaly Detection as an Enabler for Securing Network and Service Management Automation in Beyond 5G Networks

“…In addition to security, there are also AI models in the proposed architecture that are used for things like network resource optimization and proactive and reactive incident analysis in the domain analytic service. AI can help with service management, and Rizwan et al [50] show how to use AI-enabled CDR analysis for service management in zero-touch networks. For identifying suboptimal network performance and root causes, they used k-means clustering, support vector mechanisms, and their own algorithm.…”

Dynamic Network Slice Scaling Assisted by Attention-Based Prediction in 5G Core Network

“…Over the last few decades, machine learning and deep learning models have achieved considerable success in computer vision, natural language processing and time series forecasting. A few researchers have also proposed deep-learning based forecasting algorithms to forecast the future demand of VNFs in general cloud-based environment [16]- [21] or network slicing-based applications [6], [7], [22]- [24]. LSTM and three-dimensional convolutional neural network (3D-CNN) are two state-ofthe-art algorithms recently used to address the resources usage forecasting problem in network slicing.…”

“…The 3D-CNN model was developed from legacy CNN and 2D-CNN models by adding a time dimension to the problem and transforming the input into a 3D-tensor [7]. In [7] and [6], a 3D-CNN-based mechanism was proposed to predict the future traffic load in each network slice. These works overcome some limitations of classical statistical models to achieve high performance in short-term prediction , that is, 5 or 10 minutes ahead because these algorithms learn a sequential historical information of input data to predict the future value instead of only learning the current information as performed by other algorithms.…”

“…The preprocessed multivariate time series data x ′′ t−T :t starting from t − T to t are fitted into the attention-based encoder-decoder forecasting module F orecasting to forecast the future resources usage of VMs (5). In sequence, the output f of the attention-based forecasting module F orecasting is then sent to the scaling decision module which compares the future resources usage with thresholds to decide whether to add or remove a VNF instance (6).…”

“…Although network slicing technology can provide better performance and more efficient networks, it confronts a significant challenge in resource management and scaling [6]. Network operators have to ensure that the Service Level Agreement (SLA) requirements for latency, bandwidth, resource, etc for each network slice are guaranteed despite the variability of the end-user requests arrivals in each slice.…”