Deep Learning Based Fusion Model for Multivariate LTE Traffic Forecasting and Optimized Radio Parameter Estimation

Abstract

With the evaluation of cellular network internet data traffic, forecasting and understanding traffic patterns become the critical objectives for managing the network-designed Quality of Service (QoS) benchmark. For this purpose, cellular network planners often use different methodologies for predicting data traffic. However, traditional traffic forecasting approaches are erroneous. As well as most of the time, traditional traffic forecasts are high-level or a generously large regional cluster level. Also, eNodeB-level utilization with concerning traffic forecasting is not readily available. As a result, user experience degradation or unnecessary network expansion is triggered based on the traditional method. This research deals with extensive 6.2 million real network time series Long-Term Evolution (LTE) data traffic and other associate parameters, including eNodeB-wise Physical Resource Block (PRB) utilization, which focuses on building a traffic forecasting model with the help of multivariate feature inputs and deep learning algorithms. A stateof-the-art Deep Learning algorithm-based fusion model is proposed. The combination of different deep learning algorithms, namely Long Short-Term Memory (LSTM), Bidirectional LSTM (BiLSTM) and Gated Recurrent Unit (GRU), enables traffic forecasting at a granular eNodeB-level and also provides eNodeB-wise forecasted PRB utilization. In this research R 2 score value for the proposed fusion model is 0.8034, which outperforms traditional models. Apart from the PRB utilization, QoS threshold was devised as 70% from a real network experience to trigger soft parameter tuning decisions. Based on the forecasted PRB utilization, this research proposed a unique algorithm that estimates eNodeB-level soft capacity parameter optimization for a short-term step-up solution or long-term network expansion to ensure a guaranteed QoS benchmark.

Improving Heart Disease Probability Prediction Sensitivity with a Grow Network Model

“…A wide range of performance matrices [26] are used to analyze and compare the ability of the proposed prediction model. The balanced dataset refined through the Synthetic Refinement Pipeline was used in this section.…”

“…Equations to compute the specificity [26] and sensitivity [27]: Specificity indicates how well a model correctly identifies those who don’t have a probability of heart disease, and sensitivity shows how well a model correctly identifies those who have a probability of heart disease. The obtained specificity and sensitivity across different approaches are shown in Table III .…”

DBSTGNN-Att: Dual Branch Spatio-Temporal Graph Neural Network with an Attention Mechanism for Cellular Network Traffic Prediction

“…There are many evaluation indexes to evaluate the forecasting effect of the model [36][37][38][39][40], such as the mean absolute error (MAE), root-mean-square error (RMSE), mean absolute percentage error (MAPE), mean squared error (MSE), and R-squared (R2). In this paper, the three most widely used evaluation indicators MAE, RMSE and R 2 were used.…”

Machine Learning-Based Cellular Traffic Prediction Using Data Reduction Techniques

“…Moreover, clustering techniques are applied to large datasets, so the data is partitioned into clusters containing similar elements. Then, an extraction rule is estimated based on the pattern of occurrence of data tokens [12], [13]. In addition, for unknown traffic, dividing data into classes requires more information on the nature of the traffic.…”