The expanding interrelated digital systems across the board have significantly increased the number and advanced nature of cyber threats, thus making traditional rule-based security approaches inadequate. This paper proposes a secure and intelligent framework that is used for detecting cyber threats using machine learning techniques. The proposed framework supplements the above-mentioned techniques through a well-planned combination of data preprocessing, discriminative feature selection, ensemble-based supervised learning, and adaptive threat classification, which accurately identifies the malicious activities in the complex network environments. By taking advantage of machine learning models that were trained using benchmark datasets of intrusion detection tests, the framework was able to differentiate between normal and malicious traffic highly effectively, while minimizing false alarms. Experimental results show that the proposed method outperforms the conventional machine learning classifiers based on the results of accuracy, recall, F1-score, and false alarm rate. The framework further includes adaptability (periodic retraining) that facilitates resilience to changing patterns of threats. Owing to its modular design, scalability, and powerful detection capabilities, the proposed solution is suitable to be implemented in enterprise networks, cloud infrastructures, and other real world cybersecurity networks.