Ransomware has become a widespread and destructive cyber threat which is capable to encrypt vital information and disable systems. Conventional signature-based detection approaches are dependent on known signatures, becomes less useful in the face of the changing environment of zero-day attacks and polymorphic malware. This paper presents a novel approach for detecting ransomware that utilizes sophisticated behavioural analysis and machine learning algorithms. The system merges intelligent feature elimination using Recursive Feature Elimination (RFE) to enhance model performance and efficiency. The experimental results using UGRansome dataset shows that the Random Forest model achieved a high accuracy of 99.50% after feature selection which outperforms the baseline model. The addition of Explainable AI (XAI) using SHapley Additive Explanations (SHAP) further deepened the understanding of which behaviors are most critical for detection, making the model’s actions clear and accessible to human experts. This research truly highlights the potential of integrating smart data analysis with explainable AI to build highly effective, interpretable and trustworthy ransomware detection systems. By supporting more reliable protection of critical services and sensitive data, the proposed approach also strengthens the foundations of secure digital governance.