<p>Peer-to-peer (P2P) lending platforms are redefining finance by transforming the interactions between the lenders and borrowers. Accurately assessing credit risk and predicting defaults in P2P platforms remains a key challenge, as traditional models often fail to capture network-based characteristics. The proposed approach, IF-Net, combines Isolation Forest outlier detection and network centrality metrics to enhance default prediction. Through this process, the framework uncovers the underlying similarities between the borrowers in the inlier and outlier segments. Degree, Strength, and Louvain Community are the network centrality metrics that capture the underlying borrower similarities in the loan network. The proposed model is evaluated using Decision Tree, XGBoost, and CatBoost algorithms, showing that the inclusion of centrality metrics increases accuracy and other metrics without data balancing. Shapley Additive Explanations (SHAP) interpretability analysis shows that all centrality measures consistently rank within the top five features for inlier borrowers, but only one ranks highly in the outlier group. Using the Lending Club dataset, this research shows that network centrality metrics improve default risk prediction, achieving a 4.6% higher accuracy for inliers compared to outliers. This finding demonstrates that graph-based metrics behave differently for inliers versus outliers, and by leveraging these differences, the model improves credit scoring accuracy.</p>