Machine learning model for performance prediction in mobile network management

Nowadays, mobile network management involves human interaction and human work ranging from conducting drive tests that can evaluate network performance and coverage to diagnosing customer complaints. Predictive network analytics related to network management involves predicting network performance at locations or times in which no direct measurement data is available. One of the major challenges when applying machine learning is to choose the best model from a variety of models to solve a problem. Therefore, the research gap is proposing the best machine learning model for predicting mobile network performance in the identified dimensions. The methodology includes drive test measurement for data collection, exploratory data analysis, data preparation, and applying machine learning models in predicting mobile network performance. Whereas throughput has a strong correlation with signal strength, throughput is the targeted parameter in network performance prediction. Three machine learning models were applied in this study which are Random Forest, Gaussian Process Regression, and K-Nearest Neighbor for throughput prediction. Based on the results and analysis of the evaluation metric comparison, it shows that the Random Forest model comes with the highest performance prediction with the R2 score of 0.79 followed by KNN 0.66 and lastly Gaussian Process Regression 0.34. Random forest achieved the best result because of an additional layer on randomness that can lessen the variance thus increasing the model accuracy. Using the hyperparameter tuning the number of trees and the value for the depth of each tree in the forest will increase random forest model accuracy. Based on the important features, the location of the measurement and SNR value is important feature in affecting network performance. Thus, network operators need to improve the network coverage in a certain area to give a better experience to the user.