A Novel Anti-classification Approach for Knowledge Protection

Area of Science:

• Computer Science
• Data Security
• Machine Learning

Background:

• Classification is crucial in data analysis, with widespread applications including medical science.
• Protecting classification knowledge is increasingly important due to the rise of cloud computing environments.
• Existing methods for knowledge protection may lack effectiveness against sophisticated attacks.

Purpose of the Study:

• To propose a novel algorithm, Shaking Sorted-Sampling (triple-S), for protecting classification knowledge.
• To enhance the security of datasets used in classification tasks, especially within cloud environments.
• To evaluate the effectiveness of the proposed algorithm against common classification models.

Main Methods:

• The triple-S algorithm sorts data using principal component analysis to ensure feature similarity between adjacent data points.
• Generated noise data with incorrect classes is introduced into the original dataset.
• An effective positioning strategy is developed to ensure accurate removal of noise data, preserving the original dataset.

Main Results:

• The disturbance effect of the triple-S algorithm on CLC, MySVM, and LibSVM classifiers increases with the ratio of noise data.
• The triple-S algorithm demonstrates a more significant disturbance effect on MySVM and LibSVM compared to existing methods, especially at higher noise data ratios.
• The proposed method effectively protects classification knowledge against unauthorized access or modification.

Conclusions:

• The Shaking Sorted-Sampling (triple-S) algorithm provides a robust method for protecting classification knowledge in datasets.
• The algorithm's effectiveness is validated through experimental results showing increased disturbance with higher noise ratios.
• Triple-S offers a promising solution for enhancing data security in machine learning applications within cloud environments.