Machine Learning-Based Cellular Traffic Prediction Using Data Reduction Techniques

Estimating and analyzing traffic patterns is essential for managing Quality of Service (QoS) metrics in cellular networks. Cellular network planners often employ various approaches to predict network traffic. However, existing algorithms rely on large datasets, leading to significant time complexity and resource demands. To address this issue, we introduce a novel algorithm, AML-CTP (Adaptive Machine Learning-based Cellular Traffic Prediction), which is trained on a small, accurate dataset to enhance prediction accuracy while reducing complexity. Our methodology includes data normalization using the Min-Max Scaler, feature selection via the Select-K-Best algorithm, and dimensionality reduction through PCA. We apply density-based clustering techniques (DBSCAN and Kernel Density) to identify high-similarity clusters for training. We evaluate several machine learning algorithms, including Support Vector Machine (SVM), Linear Regression, Decision Tree, Light Gradient Boosting, and XGBoost, using a Cellular LTE dataset from an Egyptian company. The results demonstrate that the Decision Tree algorithm achieved the highest R² score of 96%, followed by the extension XGBoost model, which reached a remarkable R² score of 98%, indicating its superior performance in cellular traffic prediction.