Machine Learning-Assisted Mobile Network Speed Prediction: A Real-World Case Application

In recent times, as the use of mobile devices has significantly increased, there is a growing expectation for uninterrupted and high-quality wireless mobile network services in all areas. In line with this expectation, ongoing performance measurements and analyses are conducted. However, conducting such studies in an area where no measurements have been taken would result in a loss of both cost, increased workload, and time. To prevent such losses, machine learning was developed to predict the speed test results based on data collected previously with traditional measurement devices. This machine learning model has been designed to anticipate the speed test outcomes in an area where no measurements have yet been taken, relying on the similarity to previously measured parameters. In this study, a variety of data from different sources has been used. Various parameters such as different cities, residential areas, mobile device brands, and various generations of mobile networks were utilized as inputs in the machine learning modeling. The performance metric of the mobile network speed test is used as output. Various models were applied to this dataset. The achieved results were compared, leading to the selection of the algorithm that provided the most suitable results for this study. The results overall suggest that the Random Forest Regression model generally performed better in predicting mobile network performance in these cities, while AdaBoost and XGBoost had varying degrees of accuracy and predictive power in different regions. The study can be used in service quality prediction for internet providers by integrating relevant machine learning models into real systems with application programming interfaces (API).