Anti-Classification –A Kind of Anti-Data Mining
| Autor: | Yung-Ching Lin, 林永慶 |
|---|---|
| Rok vydání: | 2008 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 96 Data Mining is used to mine meaningful and relational information from large database. The information can be used by decision makers to plan strategies. The information can be used by enterprises to construct effective predictors, classifiers or rules. Competitors or users with malicious intents who mined information illegally might discover the hidden important information or trend. The information could be used by competitors to make decisions that resulted in a great loss to enterprise in losing its competitiveness. In the past, the main objective of database security is to protect data integrity, availability, and confidentiality of data; thereby, guarding sensitive and private data from being accessed by illegal users with hostile intents to steal, modify, or break the data. Current database security techniques are ineffective when enterprises had to give access to manager, but still need to prevent important information from being mined by these managers. That is, once access had been given there is no way to prevent mining of important and confidential information from the database. In this paper, we propose a new data security concept Anti-Data Mining (ADM). The objective of ADM is to prevent illegal users from adopting data mining methods to analyze useful information from data warehouse. Enterprises can adopt ADM to effectively prevent illegal users from mining useful information, trend or pattern from databases which might reveal the future development direction of enterprises; thereby, enabling enterprises to maintain their competitive ability. The proposed ADM is focused on Anti-Classification (ACF) that protects each classification algorithm effectively by the “Shaking Random Sampling Interference Algorithm (SRSI)”. Experimental results show that after producing the shake random sampling data to upset each classification algorithm in order to induce wrongly classified model that can decrease prediction accuracy. The proposed method should have the following advantage: (1) users could access any data of database normally, (2) illegal users have difficulty differentiating the correct data and the interference data, (3) user can manipulate parameters based on demands to reduce estimate accuracy or limit the ratio of interference set to data, and (4) users can recover to the original database from the anti-database with the correct parameter |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|