PrivacyAccountabilityTrustworthinesscs.CR
Privacy-Aware Machine Unlearning with SISA for Reinforcement Learning-Based Ransomware Detection
Jannatul Ferdous, Rafiqul Islam, Md Zahidul Islam
April 18, 2026
The Web Conference
3 citations
Key Takeaways
-
Machine unlearning enables efficient "right to be forgotten" compliance: The paper demonstrates that organizations can remove the influence of specific training data from ransomware detection systems without retraining from scratch, addressing privacy regulations like GDPR that require data deletion capabilities.
-
Sharding approach reduces computational burden of data deletion: By partitioning training data into separate model sections (shards), the framework allows retraining only the affected shard when data must be removed—rather than rebuilding the entire system—while maintaining detection performance with less than 0.05% accuracy loss.
-
Privacy protections don't require sacrificing security performance: The study found that implementing machine unlearning in ransomware detection systems preserved detection effectiveness, suggesting that responsible AI practices and cybersecurity don't have to be in tension with each other.
-
Responsible AI deployment is becoming technically feasible: This work demonstrates that privacy-aware machine learning isn't just a theoretical concept but can be practically implemented in real security applications, making it increasingly viable for organizations to meet both performance and ethical standards.
Abstract
Ransomware detection systems increasingly rely on behavior-based machine learning to address evolving attack strategies. However, emerging privacy compliance, data governance, and responsible AI deployment demand not only accurate detection but also the ability to efficiently remove the influence of specific training samples without retraining the models from scratch. In this study, we present a privacy-aware machine unlearning evaluation framework for reinforcement learning (RL)-based ransomware detection built on Sharded, Isolated, Sliced, and Aggregated (SISA) training. The framework enables efficient data deletion by retraining only the affected model shards rather than the entire detector, reducing the retraining cost while preserving detection performance. We conduct a controlled comparative study using value-based RL agents, including Deep Q-Network (DQN) and Double Deep Q-Network (DDQN), under identical experimental settings with a cost-sensitive reward design and 5-fold cross-validation on Windows 11 ransomware dataset. Detection confidence is evaluated using a continuous Q-score margin, enabling ROC-AUC analysis beyond binary predictions. For unlearning, the dataset is partitioned into five shards with majority-vote aggregation, and a fast-unlearning path is evaluated by deleting 5% of the samples from a single shard and retraining only that shard. Results show that SISA-based unlearning incurs negligible utility degradation (<= 0.05 percent F1 drop) while substantially reducing retraining time relative to full SISA retraining. DDQN exhibits slightly improved stability and lower utility loss than DQN, while both agents maintain near identical in-distribution performance after unlearning. These findings indicate that SISA provides an efficient unlearning mechanism for RL-based ransomware detection, supporting privacy-aware deployment without compromising security effectiveness.