AI-POWERED RANSOMWARE MITIGATION: AUTOENCODER-BASED GNN FOR EARLY THREAT DETECTION AND RESPONSE

Ransomware attacks represent significant cybersecurity headaches with serious financial implications and operationaldisruption. Traditional security solutions, including antivirus software and rule-based intrusion detection systems, arebeing hampered by the growing sophistication of ransomware threats. There is a clear need for AI-based earlydetection and mitigation solutions that operate in real time. This work proposes an Autoencoder-Based GNNframework for proactive ransomware detection. Autoencoders are used by the model to learn the patterns of normalbehaviour in the network and alert deviations from those behaviours related to ransomware attacks. The GNN supportsthe analytical processing by capturing convoluted interrelationships within the network, which in turn improves theaccuracy of threat detection. Using Wireshark to collect the dataset, and preprocessing involves Min-Max scaling fornormalization of features. The VAE then encodes the network traffic into a latent space, deviations from which are themeasures of anomaly. Anomalies distinguished by a thresholding mechanism will classify traffic into benign versusransomware. Experimental results show an increase in accuracy by 14% over time: starting from 84% during the firstmonth and reaching 98% thereafter within the twelfth month. It was shown that the model surpassed otherconventional security systems achieving high detection rates while keeping false-positive rates low. The results showthat a rapid AI response may reduce the impact of ransomware significantly; hence, there is a need for continuouslearning cyber security frameworks. Future work will proceed with the reinforcement learning-based hybrid model tofurther improve the mitigation of threats.