Link prediction is critical for understanding the changing structure of complex networks and having applications in social network analysis, recommendation systems, and biological networks. Traditional approaches relying on heuristic similarity measures often fail to capture the intricate structural dependencies within graphs. To address this challenge, we propose LPKatzn2vec-XGBoost, a novel link prediction framework that integrates Katz similarity matrix, node2vec (n2vec) embeddings, and the XGBoost ensemble classifier to enhance predictive performance. The proposed approach first constructs a Katz similarity matrix to capture higher-order connectivity patterns by considering all possible paths between node pairs. These similarity scores are then utilized in the node2vec embedding model, which preserves both local and global structural information in a low-dimensional vector space. The resulting node representations are transformed into edge embeddings and fed into XGBoost, a powerful ensemble learning classifier, to predict potential links. It effectively balances positive and negative edge samples, ensuring robust and unbiased classification. The results validate the effectiveness of integrating higher-order similarity measures with self-supervised embedding techniques and ensemble learning for improved link prediction performance.