Abstract

A data classification system is designed consisting of three layers. The second layer is the main focus of this research paper. It describes a meta-learning (learning to learn) concept that uses certain characteristics of the dataset as well as some more general knowledge about supervised and unsupervised machine learning algorithms (e.g. supervised learners tend to perform very well in the presence of a large pre-labelled training sets, etc.) to create some hypothesis. The main aim of this research is to harness general knowledge about a dataset and different machine learning methods to develop a set of meta-rules that when implemented will help to automate and speed up big data classification processes in data mining. An experiment is conducted to verify the hypotheses made using supervised and unsupervised knowledge flows in weka with some datasets taken from weka and UCI machine learning repositories. The performance result of the experiments is used to design a meta-learning algorithm in form of rules. The results from the experiments confirmed that general knowledge known about supervised and unsupervised learning is then harnessed successfully for making learning decisions.

Recommended Citation

Ighoroje, L., Lu, J., & Xu, Q. (2016). Hybrid classification system design using a decision learning approach and three layered structure - A Meta learning paradigm in Data Mining. In J. Gołuchowski, M. Pańkowska, C. Barry, M. Lang, H. Linger, & C. Schneider (Eds.), Information Systems Development: Complexity in Information Systems Development (ISD2016 Proceedings). Katowice, Poland: University of Economics in Katowice. ISBN: 978-83-7875-307-0. http://aisel.aisnet.org/isd2014/proceedings2016/CogScience/8.

Paper Type

Event

Share

COinS
 

Hybrid classification system design using a decision learning approach and three layered structure - A Meta learning paradigm in Data Mining

A data classification system is designed consisting of three layers. The second layer is the main focus of this research paper. It describes a meta-learning (learning to learn) concept that uses certain characteristics of the dataset as well as some more general knowledge about supervised and unsupervised machine learning algorithms (e.g. supervised learners tend to perform very well in the presence of a large pre-labelled training sets, etc.) to create some hypothesis. The main aim of this research is to harness general knowledge about a dataset and different machine learning methods to develop a set of meta-rules that when implemented will help to automate and speed up big data classification processes in data mining. An experiment is conducted to verify the hypotheses made using supervised and unsupervised knowledge flows in weka with some datasets taken from weka and UCI machine learning repositories. The performance result of the experiments is used to design a meta-learning algorithm in form of rules. The results from the experiments confirmed that general knowledge known about supervised and unsupervised learning is then harnessed successfully for making learning decisions.