A Strategy to Develop and Translate Custom 5G Network Slice Templates Using Machine Learning Techniques

5G networks will support customization through network slicing technology. The design and planning of network slices to meet disparate demands from divergent use cases is necessary but complex. This task involves mapping and aligning network performance metrics to application services network demands in order to design custom network slice templates. On the other hand, 5G is anticipated to generate a lot of data. When this data is processed, useful insights such as network traffic patterns can be produced to aid the design of customized network slices. Machine learning techniques present an opportunity to process this data. As a result, this paper proposes a five-step strategy augmented by machine learning techniques on existing and futuristic 5G network-related datasets to aid the design of custom network slice templates. The strategy involves, data pre-processing and dimensionality reduction, clustering, classification, and feature selection tasks. Afterward, this work explains a mapping or translation of the outputs from the machine learning tasks to possible customized network slice template attributes defined by the technical standards. Experimentation and preliminary results from step one of the strategy, data pre-processing and dimensionality reduction, indicate that one-hot encoding outperforms ordinal encoding as an encoding scheme when combined with dimensionality reduction techniques as well as other subsequent machine learning tasks proposed in the strategy.