Abstract:This paper proposes an interpretable user-behavior-based (UBB) network traffic prediction (NTP) method. Based on user behavior, a weekly traffic demand profile can be naturally sorted into three categories, i.e., weekday, Saturday, and Sunday. For each category, the traffic pattern is divided into three components which are mainly generated in three time periods, i.e., morning, afternoon, and evening. Each component is modeled as a normal-distributed signal. Numerical results indicate the UBB NTP method matches the practical wireless traffic demand very well. Compared with existing methods, the proposed UBB NTP method improves the computational efficiency and increases the predictive accuracy.
What problem does this paper attempt to address?
This paper attempts to solve the problem of network traffic prediction (NTP). Specifically, the paper proposes an interpretable network traffic prediction method based on user - behavior modeling (User - Behavior - Based Network Traffic Prediction, UBB NTP). The following are the main problems and solutions in the paper:
### Research Background
With the rapid development and popularization of Internet technology, the demand for network traffic has increased dramatically. By 2025, there will be 5.7 billion Internet users and 32.1 billion connections worldwide, which will generate a large amount of data. Under the combined influence of large - scale data, heterogeneous networks, and terminal mobility, network efficiency faces severe challenges. Therefore, network traffic analysis and prediction techniques are considered one of the important methods to improve network efficiency.
### Shortcomings of Existing Methods
Existing network traffic prediction methods are mainly divided into two categories: statistical - based methods and machine - learning - based methods. Statistical - based methods such as ARMA and ARIMA models perform well in extracting linear features from historical data, but have limitations in prediction accuracy and computational efficiency. Machine - learning - based methods, especially deep neural networks (DNN), perform excellently in prediction accuracy, but the selection of hyper - parameters is complex and time - consuming, and lacks interpretability.
### Proposed Method
The paper proposes a user - behavior - based network traffic prediction method (UBB NTP), which has the following characteristics:
1. **Interpretability**: By regarding network traffic as the superposition of user behaviors, the method provides visual and easy - to - understand prediction results.
2. **Computational Efficiency**: Compared with existing methods, the UBB NTP method has significant advantages in computational efficiency.
3. **Correspondence between Parameters and User Habits**: The model parameters correspond to user habits and can be used to compare traffic patterns in different regions.
### Specific Implementation of the Method
1. **User Behavior Analysis**: Divide a day into three main periods (morning, afternoon, and evening), and the traffic in each period is regarded as a normal - distribution signal. It is assumed that users' traffic demands reach their peaks during these periods, and due to various subjective or objective factors, the traffic demands of some users will deviate from the peak time.
2. **Traffic Data Analysis**: Use short - message - service (SMS) datasets from Guangzhou, China, and Milan, Italy, to extract traffic patterns. Traffic data shows periodic changes within a week, which conforms to the natural cycle of human activities.
3. **Mathematical Model**: Decompose the daily traffic curve into traffic components of three main periods, and each component is represented by a normal - distribution signal. Estimate the model parameters through an optimization algorithm to minimize the error between the predicted value and the actual value.
### Experimental Results
1. **Prediction Accuracy**: The experimental results show that the UBB NTP method is superior to existing statistical methods and machine - learning methods in prediction accuracy, especially in indicators such as mean - squared error (MSE), root - mean - squared error (RMSE), mean - absolute error (MAE), and coefficient of determination (R²).
2. **Computational Efficiency**: The UBB NTP method is significantly superior to other methods in computational efficiency, especially deep - learning methods. For example, on the Guangzhou dataset, the total training and prediction time of the UBB NTP method is only 8.7 seconds, while the LSTM network requires about 242 seconds.
### Conclusion
The paper proposes a user - behavior - based network traffic prediction method (UBB NTP), which is superior to existing methods in both prediction accuracy and computational efficiency, and has good interpretability. Future research will explore the similarities and differences of different types of traffic data, develop seasonal models to predict long - term traffic trends, and further optimize parameter selection to improve prediction accuracy and computational efficiency.