For A Better or Worse: Evaluating the Hybrid Feature Selections in Predicting Mobile Network Performance

Today, the proliferation of smart devices and mobile networks, alongside activities like social networking, online gaming, and video streaming, has led to the generation of vast amounts of data. This surge in data consumption has placed significant pressure on mobile service providers to deliver higher data throughput to meet growing demands. As a result, mobile operators require efficient feature selection strategies to optimize throughput while ensuring the effective use of network resources. Feature selection is critical in improving network performance by identifying and prioritizing key parameters that significantly influence throughput. This paper introduces a hybrid feature selection approach that combines mutual information as a filter-based method with Recursive Feature Elimination using an Extra Tree Regressor as a wrapper-based method. The selected features are evaluated using three machine learning algorithms: Extra Tree Regressor, Random Forest, and Extreme Gradient Boosting. Experimental results indicate that the proposed feature selection method, when paired with the Extra Tree Regressor, outperforms both Random Forest and Extreme Gradient Boosting in terms of Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and the R-squared (R²) metric.