An evolutionary computation-based sensitive pattern hiding model under a multi-threshold constraint in healthcare.

Area of Science:

• Data Mining
• Privacy-Preserving Techniques
• Machine Learning

Background:

• Collaborative frequent pattern mining faces privacy challenges due to sensitive data.
• Existing methods like Particle Swarm Optimization (PSO) and Ant Colony Optimization (ACO) often reduce data utility by deleting transactions or items.
• This can lead to significant side effects and loss of valuable information.

Purpose of the Study:

• To propose a novel Particle Swarm Optimization (PSO) based algorithm for enhanced privacy preservation in collaborative frequent pattern mining.
• To improve the utility of sanitized datasets by selecting victim items based on multiple parameters.
• To introduce a dynamic multi-threshold framework for more effective sensitive pattern concealment.

Main Methods:

• A novel Particle Swarm Optimization (PSO) algorithm is developed for sensitive pattern concealment.
• A dynamic multi-threshold framework is introduced, utilizing bi-variate normal distribution to set dynamic thresholds.
• Victim item selection is based on multiple parameters to minimize data modification and preserve utility.

Main Results:

• The proposed PSO algorithm effectively conceals sensitive patterns across benchmark datasets (FIMI, Heart Disease, Heart Attack Prediction).
• It significantly reduces side effects and data loss compared to existing PSO and Ant Colony Optimization (ACO) based algorithms.
• The Failure-to-Hide (FTH) metric is notably lower, indicating superior privacy preservation.

Conclusions:

• The novel PSO algorithm offers a superior approach to privacy-preserving data mining.
• It effectively balances data privacy with data utility, outperforming existing methods.
• The dynamic multi-threshold framework enhances the applicability of privacy-preserving techniques in real-world scenarios.