Leukemia, a complex and heterogeneous group of blood cancers, presents significant diagnostic and prognostic challenges due to its genetic diversity. Gene expression profiling has emerged as a powerful approach for leukemia subtype classification, yet the interpretability of machine learning models built on such data remains a pressing concern in clinical settings. This study introduces a novel SHAP-driven machine learning framework that leverages Synthetic Minority Oversampling Technique (SMOTE) to handle class imbalance and employs SHapley Additive exPlanations (SHAP) to interpret model predictions. Using the publicly available Golub et al. Dataset, the proposed framework demonstrates high classification accuracy while offering interpretable insights into gene contributions toward disease prediction. The integration of SHAP enables clinicians and researchers to better understand model behavior by identifying the top 10 key biomarkers that aid the classification, and thus support explainable decision-making in oncology. This research thus contributes to the growing field of Explainable Artificial Intelligence (XAI) and marks a significant step toward transparent and trustworthy AI applications in leukemia diagnosis and precision medicine.