Downlink throughput prediction using machine learning models on 4G-LTE networks

• Published in: International journal of information technology (Singapore. Online) 2023-08, Vol.15 (6), p.2987-2993
• Main Authors: Al-Thaedan, Abbas, Shakir, Zaenab, Mjhool, Ahmed Yaseen, Alsabah, Ruaa, Al-Sabbagh, Ali, Salah, Monera, Zec, Josko
• Format: Article
• Language: English
• Subjects: Artificial Intelligence; Computer Imaging; Computer Science; Image Processing and Computer Vision; Machine Learning; Original Article; Pattern Recognition and Graphics; Software Engineering; Vision
• ISSN: 2511-2104; 2511-2112
• DOI: 10.1007/s41870-023-01358-9

With the enormous evolution of the smartphone, especially with the appearance of the fourth generation (4G) cellular networks, the demand for high-speed data rate, low latency, and video streaming have been increased. This rising demand for network utilization has demonstrated the need for more service improvement. Furthermore, with rising demand and complexity, traditional network management techniques are inadequate, necessitating an autonomous calibration to reduce system parameter usage and processing time. Therefore, real network monitoring and performance analysis should be applied by utilizing various models. Because Downlink Throughput (DL-Throughput) holds significant importance factors for network performance, DL-Throughput prediction can be used to evaluate the quality of cellular networks. Various Machine Learning (ML) models utilized Long-Term Evolution (LTE) data measurements for the prediction process. In this article, the selected ML models Support Vector Regression (SVR), Linear Regression (LR), K Nearest Neighbors (KNN), and Decision Tree Regression (DTR) have been used for forecasting DL-Throughput from three different cellular network operators in an urban area. The parameters with high correlation on throughput and are used as feature selection with ML are the GPS coordinates, RSRP, RSRQ, SINR, and RSSI. The statistical analysis has been utilized to determine the accuracy of the ML models. As a result, the KNN and DTR obtain the best accuracy in the three operators compared with other ML models. For instance, the accuracy for R 2 of DTR is 99%, 93%, and 98% with operator 1 (OPR1), operator 2 (OPR2), and Operator 3 (OPR3), respectively.