Critical infrastructure, such as power grids and healthcare systems, faces rising cyber threats like ransomware and APTs, which traditional security measures fail to detect, requiring an AI-driven approach. This research introduces an MLTIF that integrates structured and unstructured threat intelligence sources with machine learning-based classification models to enhance threat detection, response, and resilience. Data Collection and Integration aggregates logs from SIEM systems, ICS monitoring, OSINT, dark web monitoring, and commercial threat feeds. Data Preprocessing and Feature Extraction standardize data, remove noise, and extract key security attributes such as network anomalies and unauthorized access patterns. Machine Learning-Based Threat Detection employs ANN, Random Forest, KNN, and anomaly detection algorithms to classify cyber threats. Real-Time Threat Mitigation automates responses by isolating compromised systems, blocking malicious IPs, and alerting security teams. An Adaptive Learning and Feedback Mechanism continuously updates the AI-models to enhance detection capabilities. Experimental results show that Artificial Neural Networks achieved a detection accuracy of 99.2%, outperforming other models such as AdaBoost with 98.65%, Decision Tree with 97.8%, and KNN with 97.9%. The framework successfully detected 99.5% of denial-of-service attacks, 98.6% of SQL injections, and 97.8% of phishing attempts. Real-time mitigation strategies reduced incident response time by 40%, while the adaptive feedback loop improved detection rates. Cloud-based environments showed the highest detection accuracy of 99%, while IoT networks exhibited the lowest performance at 96.8% due to resource constraints. The proposed framework ensures real-time, adaptive, and automated cyber defense, significantly improving critical infrastructure security and operational resilience.