PriMa$\chi$: Novel Genetic Reinforcement Learning Model for Improving Privacy Preservation in Blockchain

Abstract

As privacy concerns intensify in data-driven systems, this article presentsPriMa$\chi $, a hybrid framework that combines a genetic algorithm (GA) and reinforcement learning (RL) to optimize the privacy–utility tradeoff in differential privacy (DP) through explicit adaptive privacy–utility control. PriMa$\chi $adaptively selects perturbation configurations to minimize the privacy budget$(\varepsilon)$while preserving data utility$({\mathcal {U}})$. To support verifiable privacy-preserving analytics in decentralized environments, we further integrate PriMa$\chi $with a privacy-aware smart-contract framework that enables on-chain DP enforcement and zero-knowledge proof (ZKP) verification. The framework supports structured, transactional, and spatiotemporal workloads, including decentralized finance, electronic health records, census analytics, and location services. An interleaved Petri net model is used to formally verify privacy-aware state transitions in the smart-contract workflow. Experimental results show that PriMa$\chi $achieves utility of at least 80% under dataset-dependent privacy budgets in the range$0.003 \leq \varepsilon~\leq\unicode{0x0142}.43$, while also effectively mitigating model-extraction, membership-inference, and privacy-budget-exhaustion attacks. These results demonstrate that PriMa$\chi $provides adaptive, auditable, and practically deployable privacy protection for decentralized analytics.