Survey of Multi Relational Classification (MRC) Approaches & Current Research Challenges in the field of MRC based on Multi-View Learning
Amit Thakkar1, Y P Kosta2
1Amit Thakkar, Department of Information Technology, Charotar Institute of Technology (Faculty of Technology and Engineering), Charotar University of Technology Changa, Anand, Gujarat, India.
2Y P Kosta Marwadi Education Foundation’s Group of Institutions, Rajkot, Gujarat, India.
Manuscript received on December 05, 2011. | Revised Manuscript received on December 19, 2011. | Manuscript published on January 05, 2012. | PP: 247-252 | Volume-1 Issue-6, January 2012. | Retrieval Number: F0311111611/2012©BEIESP
Open Access | Ethics and Policies | Cite
© The Authors. Published By: Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Abstract: An increasing number of data mining applications involve the analysis of complex and structured types of data and require the use of expressive pattern languages. Many of these applications cannot be solved using traditional data mining algorithms. This observation forms the main motivation for the multi-disciplinary field of Multi-Relational Data Mining (MRDM). Unfortunately, existing “upgrading” approaches, especially those using Logic Programming techniques, often suffer not only from poor scalability when dealing with complex database schemas but also from unsatisfactory predictive performance while handling noisy or numeric values in real-world applications. However, “flattening” strategies tend to require considerable time and effort for the data transformation, result in losing the compact representations of the normalized databases, and produce an extremely large table with huge number of additional attributes and numerous NULL values (missing values). As a result, these difficulties have prevented a wider application of multi relational mining, and post an urgent challenge to the data mining community. To address the above mentioned problems, this article introduces a multiple view approach—where neither “upgrading” nor “flattening” is required— to bridge the gap between propositional learning algorithms and relational databases and current research challenges in the field of Multi relational classification based on Multi View Learning.
Keywords: Multi Relational Data Mining, Propositional Learning, Multi Relational Classification, Relational Learning.