On Using Linear Diophantine Equations for in-Parallel Hiding of Decision Tree Rules

Abstract

Data sharing among organizations has become an increasingly common procedure in several areas such as advertising, marketing, electronic commerce, banking, and insurance sectors. However, any organization will most likely try to keep some patterns as hidden as possible once it shares its datasets with others. This paper focuses on preserving the privacy of sensitive patterns when inducing decision trees. We adopt a record augmentation approach to hide critical classification rules in binary datasets. Such a hiding methodology is preferred over other heuristic solutions like output perturbation or cryptographic techniques, which limit the usability of the data, since the raw data itself is readily available for public use. We propose a look ahead technique using linear Diophantine equations to add the appropriate number of instances while maintaining the initial entropy of the nodes. This method can be used to hide one or more decision tree rules optimally.

Using Minimum Local Distortion to Hide Decision Tree Rules

“…The first one is that we do not need to add new instances to the original data set, and the second is that our new heuristic can be performed in only one step with much lower computational complexity compared to solving systems of Linear Diophantine Equations. However, our previous published techniques [ 20 , 21 ] guarantee the preservation of entropy values in every node of the tree before and after the modification.…”

“…This approach is critical because the sanitized data set may be subsequently published and even shared with the data set owner’s competitors, as can be the case with retail banking [ 19 ]. We extend this work in the papers [ 20 , 21 ] by formulating a generic look ahead technique that considers the structure of the decision tree from an affected leaf to the root. The main contribution of these publications was to improve the Swap-and-Add pass by following a look ahead approach instead of the greedy approach which was previously used.…”

QSAR Models for Active Substances against Pseudomonas aeruginosa Using Disk-Diffusion Test Data

“…In the learning phase, the rules are derived (tree generation) and in an accuracy verification phase, random data taken from the training set is tested and rules are adjusted in order to decrease the tree size (tree pruning); in the end the unlabeled data points are classified with the rules thus developed and tested [70,71]. Simplicity, transparency, easiness to understand and to implement [72,73] are key advantages of the decision tree classifier. The key parameter influencing the tree's performance is its maximum depth, as it decides its complexity [74]; in our models, this parameter had values between two and four.…”

Inference Control in a Diabetes Data Set Using a Java-Based Prototype of LDH Algorithm

“…In articles [5][6][7][8], the authors proposed a series of strategies that would effectively protect against the disclosure of the sensitive classification rules. The LDH algorithm [9] was developed on the basis of the concept of preserving sensitive DT rules resulting from the use of data mining techniques.…”