Cyber threats are becoming more advanced, faster, and harder to detect using traditional rule-based systems. These older systems often miss new or complex attacks, especially those found in unstructured data like reports or alerts. In this study, we propose a system that uses Machine Learning (ML) and Natural Language Processing (NLP), along with Explainable AI (XAI), to improve threat detection. The system collects behavior-related features from network traffic logs and gathers context from documents such as security advisories. Three classifiers, Random Forest (RF), Support Vector Machine (SVM), and XGBoost, are trained using these features together. XGBoost did the best of the group, with an accuracy of 95.2% and an AUC of 0.96. To help analysts understand the results, we use SHAP and LIME to explain the predictions. This method makes things more accurate and trustworthy overall, and it can be helpful in security systems that work in real time.