ML_Engine_for_UE-VBS_Selection

A Two Stage Machine Learning Approach for 5G Mobile Network Augmentation through Dynamic Selection and Activation of UE_VBSs.

Note: Use Jupyter NbViewer to load the code if GitHub doesn't render

Please copy the links of the .ipynb files and paste it in NbViewer if GitHub fails to load it in the browser.

Helpful Documents and Notebooks

• Check out the Data used to train these models.
• Check out the two-stage machine learning model Code!
• Check out the project presentation Deck!
• Check out the details of the project in this Paper!

Abstract

The 5G cellular network is the new generation of mobile networks that focus and identifies the current and future needs of the heterogeneous devices in the context of wireless access in a dense network scenario. This research builds on the user equipment-based virtual base station (UE-VBS) concept, which utilizes smartphones to provide base station services to other UEs in the perimeter by capitalizing the capabilities of the new UE in terms of massive connectivity and the enhanced resources such as computing, and battery-power it offers. However, in a dynamic network architecture like the 5G network, the selection of a qualified UE to become a UE-VBS is a challenging task with the introduction of newer technologies, UE hardware configurations, and network infrastructures. In this context, to automate the successful identification of a potential UE to act as a VBS is a need, a machine learning (ML) model can be employed. Hence, in this paper, we propose and explore a two-stage machine learning approach to dynamically choose the eligible UEs that will be activated on the fly as UE-VBS to support data rate expansion and improve quality of service (QoS) in locations where infrastructure is lacking, and a more agile network operation is required. Furthermore, the UEs are clustered based on their Euclidean distance using MeanShift, Affinity Propagation, OPTICS, K-Means, Spectral Clustering, Agglomerative Clustering and BIRCH unsupervised ML approaches, and the devices are categorized based on their eligibility to become a UE-VBS for the corresponding cluster using Random Forest, AdaBoost, and Gradient Boosting supervised ML classification approaches. Then, using a heuristic algorithm, we determine the optimal cluster head (CH) by exploiting the findings of our classification model. The proposed framework is simulated for a nonuniform distribution of the UEs in time and space and quantified using statistical analysis. Our simulation results demonstrate that the proposed model achieves an accuracy of around 95% using Random Forest classifier and the K-Means clustering.

Code Walkthrough

1. Simulation and Data Generator
2. Comparison of Clustering Algorithms
3. Classification
4. SMOTE
5. Validation and Testing
6. Pipelined Model, Heuristic Algorithm for Identification of Cluster Heads and Sum Rate Calculation

Result

In this paper, we delved into the implementation of a twostage machine learning approach for 5G mobile network augmentation through dynamic selection and activation of UE-VBSs. For our investigations, a dataset has been generated and pre-processed by simulating the UEs geographical coordinates as a non-uniform distribution in the time and space domain. By using the dataset generated, we trained a two-stage machine learning model to dynamically select UE-VBSs for each of the clusters around the primary BS. Initially, the UEs distributed around the primary BS were clustered into groups using the K-means as it is found to be the most appropriate for clustering with a silhouette score of 0.46. Then, a classifier to determine the eligibility of the UEs to become a UE-VBS for each cluster has been pipelined in the second stage. Furthermore, the class imbalance towards the majority class has been conquered using SMOTE before classification. Of the considered classification algorithms, our investigations based on the K-fold testing revealed that the Random forest classifier attains the highest mean accuracy score of 0.97. Finally, our proposed model has been subjected to statistical testing and analysis using SHAP and LIME. From the performance analysis carried out, we advocate that our proposed model-aided dynamic selection of a UE-VBS for each of the clusters around the primary BS is efficient as it has resulted in a better achievable data rate than random selection.

Installation Requirements

Environment

It is recommended to have a Linux or macOS development environment for convenience, although the code runs on Windows 10.
Use Anaconda to manage your packages and Python 3 (version >= 3.6.0 recommended).
It is also recommended to run the code on Jupyter Notebook.

Dependencies

With Anaconda, no need to install

• matplotlib
• scikit-learn
• numpy
• pandas
• seaborn
• scipy

Others

Remember to use conda, not pip for installing these

• missingno
• imblearn
• shap
• lime

Latex for documentation - Ubuntu

• texlive-full