COMPARISON OF COVERAGE PREDICTION THROUGH THE USE OF MACHINE LEARNING ALGORITHM WITH CONVENTIONAL PREDICTION MODELS

• Main Author: Edbert
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/85097
• Institution: Institut Teknologi Bandung

In an era where cellular communication technology has become a necessity in everyday life, 4G signal quality plays an important role in ensuring smooth and reliable connectivity. Users rely on cellular services for a variety of purposes, from basic communications to video streaming, online gaming and IoT applications. However, the main challenge is how to predict optimal signal coverage based on the distance from the transmitting tower, especially in varied environments such as dense urban areas, rural areas, and areas with complex topography. The need for reliable signals is increasing with the development of 5G technology which promises higher speeds, lower latency, and the ability to support more connected devices simultaneously The background to this research is based on awareness of the importance of a reliable 5G signal in supporting various activities, from communication to the use of internet-based applications and new technologies that can be implemented. The obstacles faced are related to the implementation and optimization of the 5G network which requires accurate and strong network coverage predictions. Therefore, this research was conducted to provide a new solution to overcome this problem by developing a machine learning-based prediction model using established 4G network data, so that it can be used in 5G technology. With the understanding that historical data from 4G networks can provide valuable information for 5G network deployment, this approach offers great potential in mobile network planning and optimization. With new technology such as machine learning (ML), signal coverage prediction models will continue to develop, allowing ML to become a new, accurate method for use in the future. The method used in this research consists of several stages. First, drivetest data was collected to obtain information about signal quality in various locations, such as urban and suburban areas. This data is then combined with data obtained from mobile operators to enrich the features needed to make predictions, improving accuracy and overall representation. Drivetest data provides real-time field measurements, while data from mobile operators includes technical information such as tower locations, operating frequencies, and antenna configurations. iv Next, several machine learning algorithms in supervised learning are selected and trained using the data that has been collected. In supervised learning, techniques such as regression, classification, and ensemble learning are used. The assessment metrics used are still regression-based even though some algorithms are taken from classification or ensemble learning techniques, because these algorithms can still be applied in a regression context. This machine learning model not only takes into account the distance from the transmitting tower but also other factors such as antenna height, type of environment (urban, suburban, rural), and even weather, all of which can affect signal quality. The process of developing this machine learning model considers various factors that influence signal quality, one of which is the distance from the transmitting tower to the receiver. By leveraging learning from existing data, these models are able to produce more accurate predictions of signal levels based on distance from transmitting towers. To increase prediction accuracy, comparisons were made with conventional methods such as Okumura-Hata and COST 231. This was done to validate predictions produced by machine learning, because these two conventional models are still able to make fairly accurate predictions on 4G networks and can be compared with data. actual results obtained from the drivetest. This comparison also helps identify the strengths and weaknesses of the machine learning models being developed, as well as providing insights for further improvements. The results of this research were implemented through creating an interactive dashboard using the Streamlit platform. This dashboard allows users to interact with research results, select areas for training and predictions, and upload CSV data for both processes. After execution, users can see results such as drivetest maps, comparison metrics between machine learning and conventional models, as well as graphic plots comparing predicted signal levels. By using this approach, this research succeeded in developing a 4G signal prediction model which is able to provide signal coverage estimates based on the distance from the transmitting tower to the receiver with a high level of accuracy. This research not only makes a significant contribution to improving the quality of current mobile services but also builds a strong foundation for future network development and optimization. In addition, the developed prediction model can be used as a reference for further research in the field of network optimization, especially in an era of technology that continues to develop rapidly.Keywords: Machine Learning, 4G Signal Prediction, Drivetest, Supervised Learning, Interactive Dashboard.