Core network traffic prediction based on vertical federated learning and split learning.

Area of Science:

• * Wireless communication networks
• * Machine learning for network management
• * Data privacy and distributed systems

Background:

• * Accurate wireless traffic prediction is crucial for efficient cellular network operations, including resource management and predictive control.
• * Centralized training methods face challenges with data transmission, delays, and privacy concerns.
• * Federated learning (FL) offers privacy-preserving collaborative training but struggles with heterogeneous data features across participants.

Purpose of the Study:

• * To develop an innovative framework for wireless traffic prediction that overcomes limitations of traditional methods.
• * To enable collaborative training of high-quality prediction models using diverse data while ensuring local data confidentiality.
• * To address statistical heterogeneity in distributed machine learning for improved model performance.

Main Methods:

• * Integration of split learning (SL) and vertical federated learning (VFL) for collaborative model training.
• * Edge clients train dimension-specific prediction models locally on their diverse traffic data.
• * A partially global model is shared among clients to mitigate statistical heterogeneity.

Main Results:

• * The proposed SL and VFL framework enables collaborative training of robust wireless traffic prediction models.
• * The method effectively maintains raw data confidentiality at the local edge client level.
• * Experimental results on real-world datasets demonstrate superior performance compared to existing approaches.

Conclusions:

• * The novel framework offers a privacy-preserving and effective solution for wireless traffic prediction.
• * The approach successfully handles statistical heterogeneity in distributed network data.
• * This method shows significant potential for enhancing intelligent cellular network operations through accurate forecasting.