Link Prediction in Complex Networks Using Recursive Feature Elimination and Stacking Ensemble Learning

Area of Science:

• Network analysis and modeling
• Machine learning applications
• Data science

Background:

• Link prediction is crucial for understanding network dynamics and forecasting future connections.
• Existing methods often struggle to effectively integrate diverse network topological information.
• Improving the accuracy and robustness of link prediction remains a key challenge.

Purpose of the Study:

• To propose a novel stacking ensemble framework for enhanced link prediction.
• To integrate global, local, and quasi-local network topological features.
• To improve the performance and robustness of link prediction models.

Main Methods:

• A two-level stacking ensemble model was developed for link prediction.
• Random Forest-based recursive feature elimination was used for structural feature selection.
• Base classifiers included logistic regression, gradient boosting decision tree, and XGBoost, with XGBoost in the upper level.

Main Results:

• The proposed stacking ensemble method demonstrated superior prediction results compared to existing approaches.
• Extensive experiments on six different networks confirmed the method's effectiveness.
• The approach showed significant improvements in applicability and robustness for link prediction tasks.

Conclusions:

• The novel stacking ensemble framework effectively integrates diverse network topological information for improved link prediction.
• The method offers a robust and accurate solution for predicting missing and future links in networks.
• This work contributes a valuable tool for network analysis and modeling applications.