BBO-CNN-LP: a deep learning framework for link prediction in multilayer networks using biogeography-based optimization and convolutional neural networks

Abstract Link prediction plays a crucial role in social network analysis, recommendation systems, and biological networks by identifying potential future connections based on existing network structures. While traditional heuristic-based methods such as Common Neighbors and Adamic-Adar have been widely used, they struggle with complex multilayer networks due to their inability to capture cross-layer dependencies. Recent advancements in machine learning and deep learning, including ensemble learning and graph neural networks (GNNs), have significantly improved link prediction accuracy. However, these models still face challenges related to scalability and effective feature extraction in multilayer networks. To address these limitations, this paper introduces BBO-CNN-LP, a novel deep learning-based framework for link prediction in multilayer networks. The proposed model integrates Biogeography-Based Optimization (BBO) for graph clustering, Node2Vec and Word2Vec for feature representation, and Convolutional Neural Networks (CNNs) for hierarchical learning. By leveraging both intra-layer and inter-layer dependencies, BBO-CNN-LP effectively captures structural and semantic relationships across multiple network layers. Extensive experiments conducted on the Twitter-Foursquare dataset demonstrate that BBO-CNN-LP outperforms existing models, including BEC-RLRW and traditional heuristics, in terms of accuracy, precision, recall, and AUC-ROC. These results highlight the effectiveness of deep learning-based approaches for multilayer link prediction and provide a scalable solution for real-world applications in social networks, recommendation systems, and biological research.