Abstract
dc:descriptionMy research is motivated by the emerging need for privacy-safe sharing among different organizations. In this doctoral thesis, we present two different approaches to hiding decision tree rules. In the first one, 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 adds an optimizing look-ahead perspective to an existing method, and it can be used to hide one or more decision tree rules, sequentially or in parallel. This method adopts a record augmentation approach by also modifying the class labels to hide critical classification rules in binary datasets. Moreover, we present a novel technique, Local Distortion Hiding (LDH), which does not affect the class labels of the sensitive instances, as our previous techniques do, but instead modifies the attributes’ values of these specific instances. Such a hiding methodology, as well as the first one, is preferred over other heuristic solutions like output perturbation or cryptographic techniques which they actually limit the usability of the data since the raw data itself is readily available for public use. While the new technique may need to modify more values of the initial data set, it does so by not requiring the addition of extra instances, and it thus saves on the size of the sanitized data set. This trade-off is an extra tool in the arsenal of the data engineer who might want to explore a range of possibilities when tasked with a data hiding mandate. In the proposed method, we first identify the instances that contribute to the creation of a specific rule and then, by appropriately changing attribute values, we can successfully hide this rule with minimum impact on the rest of the decision tree. Finally, we apply the method regarding the local distortion hiding (LDH) in different data sets and present some well-chosen examples, in the fields of Medicine, Finance and Education to demonstrate the efficient hiding of pre-selected decision tree rules and measure the corresponding accuracy of this method.
Degree
thesis:*- Grantor dc:publisher
- Hellenic Open University
- Year dc:date
- 2019
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Feretzakis, Georgios
- Φερετζάκης, Γεώργιος
Subjects
dc:subject × 19- Δέντρα απόφασης
- Μηχανική μάθηση
- Αλγόριθμοι τεχνητής νοημοσύνης
- Απόκρυψη δεδομένων
- Διοφαντικές εξισώσεις
- Προστασία δεδομένων
- Απόκρυψη κανόνων στα δέντρα απόφασης
- Decision tree classification
- Machine learning
- Privacy-preserving
- Data sharing
- Hiding rules
- Knowledge hiding
- Decision trees
- Local distortion hiding
- Φυσικές Επιστήμες
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Natural Sciences
- Computer and Information Sciences
Rights
dc:rights- Statement dc:rights
-
- BY_NC_SA
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/46573