PREDICTION OF 5G CELLULAR NETWORK USER MOVEMENT IN JABODETABEK AREA FOR RESOURCE AND MOBILITY MANAGEMENT USING LONG SHORT-TERM MEMORY METHOD

• Main Author: Rafinanda Benyamin, Fabian
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/85095
• Institution: Institut Teknologi Bandung

The development of communication systems worldwide has given birth to various new technologies that are rapidly developing, such as the 5G network launched in 2018. The rapid increase in the use of cellular data services demands more efficient network management. Problems begin to emerge with the increasing number of mobile cellular service users, such as signal loss, slow internet, and total network loss in certain areas. Mobile user mobility management has become an important part of user service in terms of consistent signal quality, interference reduction, and location-based services. To improve cellular services, various cellular service providers use mobility and resource management systems. This system helps in allocating resources efficiently and dynamically, increasing network capacity, and creating new more complex location-based services. Therefore, a system is needed that can predict the movement of 5G network users for the purpose of providing the best possible service. The movement prediction system will use machine learning models with various methods such as Markov Chain, Hidden Markov Model, and machine learning methods such as Support Vector Machine (SVM), XGBoost, Deep Neural Network (DNN), and Long Short-Term Memory (LSTM). LSTM is a method of creating a machine learning model that can process large amounts of sequential data, therefore the LSTM algorithm is very good for the needs of prediction systems related to time. Creating a system with an LSTM model can facilitate the prediction of the movement of 5G network users to ensure that the iv services provided are always at a high level. One way is by allocating resources to areas that are densely populated with service users. Using LSTM is quite complicated because to produce a good LSTM model, many factors determine such as the number of neurons in the layer, the number of training epochs, and the amount of data. The LSTM configuration varies depending on various factors that influence it, so it is considered a model that is quite difficult to get accurate training results. The purpose of this research is to develop an LSTM-based machine learning model that can accurately predict the movement of cellular service users over a certain period of time. The process of developing the model requires experimentation with various model layer configurations to ensure that the model used is the most accurate. The model created will receive 50 coordinate points as input and will produce a prediction of 10 points at a time. The main focus is to create an LSTM model with the right configuration so that it can accurately predict movement and help predict for the management of BTS cell resource allocation in areas that will experience an increase in density. The research results will be displayed in the form of a mapping of the trajectory along with the coordinates of the prediction results as a comparison of the accuracy level of the model prediction. The system in this research will be divided into 2, namely the prediction model and the display of prediction results. Model training uses the Grab-Position dataset consisting of millions of coordinate points originating from the travel trajectories of Grab online motorcycle taxi application users. This research will prove that the use of models with the LSTM method can predict the movement of 5G network users quite accurately and can help cellular service providers in mitigating user movement and reducing the decline in the quality of cellular network services.