Abstract: Access to formal credit remains constrained for much of the African population due to a lack of conventional credit histories. This paper investigates whether behavioral financial data mobile money transactions, airtime patterns, and savings consistency can serve as valid proxies for creditworthiness. We present FairLend-Africa, an explainable machine learning framework combining XGBoost scoring with SHAP interpretability and systematic fairness auditing. Using a synthetic dataset of 10,000 borrower records, the system achieves a held-out test ROC-AUC of 0.7137, which aligns with performance baselines established in thin-file behavioral credit scoring literature. Within the controlled synthetic data generation boundaries, the audit pipeline demonstrates no disparity under the synthetic data's demographic-behavioral independence assumption a condition requiring empirical verification with real data before deployment claims can be made. SHAP analysis identifies wallet balance trend and savings consistency as dominant signals. A logistic regression baseline matches the XGBoost performance, suggesting primarily linear structures in the synthetic data and motivating validation on real-world datasets where nonlinear interactions may emerge. The system is implemented as a REST API with an interactive dashboard released as open source to support reproducibility in African financial inclusion research. Keywords: alternative credit scoring, explainable artificial intelligence, financial inclusion, mobile money, SHAP, XGBoost, fairness in machine learning, African fintech