UTILIZATION OF MACHINE LEARNING REGRESSION MODELS FOR PREDICTING COVERAGE IN CELLULAR NETWORKS

• Main Author: Perdana Saragih, Geo
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/85093
• Institution: Institut Teknologi Bandung

The advancement of telecommunication technology continues to progress, with one significant breakthrough being the implementation of 5G cellular networks. 5G signals offer remarkable speed and capacity, yet their effectiveness relies on the deployment of appropriate infrastructure. A crucial challenge in implementing 5G networks is predicting signal coverage with high accuracy to determine the optimal locations for base station installations and to evaluate network efficiency. With the increasing complexity of urban and suburban environments, including building interference, diverse topography, and population density variations, it is essential to develop predictive methods that can address these challenges. This is particularly important to ensure the availability of reliable and quality 5G network services across various coverage areas. This research aims to develop an effective machine learning Regression model algorithm to predict signal coverage in cellular networks, identify the best-performing machine learning Regression model algorithms in predicting signal coverage in cellular networks, and analyze the performance of machine learning Regression model algorithms in predicting signal coverage in urban and suburban areas. The Regression algorithms used in this study include Linear Regression, Ridge Regression, Lasso Regression, Decision Tree Regression, Random Forest Regression, Gradient Boosting Regression, Support Vector Regression, K-Nearest Neighbors Regression dan XGBoost Regression. This research utilizes field data collected through drive tests using G-Net Track Pro and logistic cell data. This data will include various information such as signal strength, transmitter antenna height, receiver height, transmitter and receiver antenna distance, frequency, altitude, elevation angle, tilt offset angle, horizontal distance from the receiver center to the transmitter antenna axis, and vertical distance from the receiver center to the transmitter antenna axis. Regression models will be developed using this data to predict coverage in cellular networks at various locations. Model performance evaluation will be conducted using relevant methods, such as Mean Absolute Error (MAE), Mean Square Error (MSE), Root Mean Squared Error (RMSE) and R-squared (R2), to determine the most accurate machine learning Regression model algorithm in predicting 5G signal coverage in urban and suburban areas. The results show that the best-performing regression machine learning algorithm used is Random Forest Regression, with the lowest RMSE and MAE values of 2.728740 dBm and 1.724648 dBm, respectively, demonstrating Random Forest Regression's exceptional ability to minimize prediction errors. Additionally, Random Forest Regression has the best R2 value of 0.9182, indicating that the model explains 91.82% of the data variation, which is the highest achievement compared to other models. After hyperparameter tuning, Random Forest Regression remains the top model with the best RMSE and MAE values of 2.737140 dBm and 1.728487 dBm, respectively, and an R2 value of 0.917675. Although there is a slight decrease in R2 after tuning, the model shows consistent stability and accuracy, reinforcing its position as the primary choice for predicting signal coverage in cellular networks. In this study, Random Forest Regression has proven to be the best regression machine learning algorithm with significant performance in predicting cellular network coverage. Hyperparameter tuning also enhances the model's stability and accuracy, despite a slight decrease in R2. The strength of Random Forest Regression lies in its ability to handle high complexity and large data variations, as well as its good adaptation to changes in input data. The success of Random Forest Regression in this study provides a strong foundation for further research and development in the field of telecommunications, particularly in improving the quality and coverage of network services. Thus, this research not only contributes significantly to existing knowledge but also sets a standard for future studies and practical applications in the field.