5G Network Data Analysis for Development of Smart Application

Abstract

The past few years have witnessed the rapid growth of global mobile cellular traffic. After 4G newlineand now evolution toward the 5G technology, huge amount of high speed data usage is newlineexpected. With an advantage of Machine Learning, an investigation on a cellular traffic dataset newlinewith the latest deep learning models is proposed. In this work, the main aim is to identify the newlinesuitable machine learning techniques, which can be used for prediction of cellular traffic newlinedata in a Base Station. A comparative experimental study is performed among popular newlinesupervised and unsupervised learning approaches. According to the experimental analysis newlineSVM and ANN are the accurate algorithms in the supervised learning algorithms. On the other newlineside in unsupervised learning models, SOM shows better accuracy. Further as an initiative newlinetowards green energy to reduce energy consumption by switching off the Base Station servers newlineis proposed. Servers are designed to handle peak load and to work 24x7 with full capacity but newlinein actual load varies from time to time during a day leading to wastage of resources. As per newlineforecast for data utilization and categorization of Base Stations based on data usage as low, newlinemid and high a scheduling algorithm is proposed to temporarily switching off the servers during newlinenon-peak hours by this approx. 29% reduction in power consumption can be achieved. An newlineapplication by using deep learning Bi LSTM technique for finding trends and patterns of newlinehuman activities can be designed. By comparison of predicted data and actual data usage and newlineclassification based on number of users connected an estimation for crowd gathering or data newlinebroadcast in a Base Station area can be done thus enabling smart surveillance to anticipate an newlineunwanted situation by local administration. Real time dataset available in Kaggle repository newlinewere used for training and testing of ML models.