Authors: Daniel Giterman; Eyal Brill
Addresses: Faculty of Technology Management, Holon Institute of Technology, 5810201 Holon, Israel ' Faculty of Technology Management, Holon Institute of Technology, 5810201 Holon, Israel
Abstract: In real classification problems, common learning algorithms generally fail to describe instances that require complicated classification logic. Additionally, it is often difficult to ensure a satisfying amount of classified data for their training. In this work, we propose and examine a new learning algorithm that also integrates expert logic. Essentially, this algorithm takes advantage of unclassified data to produce a self-generated fuzzy inference system that is eventually used as a classifier. It also utilises a mere sample of classified data in order to compare various classifiers constructed from different algorithm options, thus finally achieving an assumingly more accurate result. As part of our study, this algorithm was compared with six well-known supervised learning algorithms such as artificial neural networks, support vector machine and random forest. We used the ten-fold cross-validation technique with Kappa statistic to assess algorithm performance. Subsequently, in order to find statistically significant dissimilarities among the algorithms, we used a two-tailed Friedman test. After the null hypothesis was rejected, we used a Nemenyi post-hoc test to prove differences between pairs of algorithms. Consequently, despite lacking in efficiency and scalability, our algorithm proved to be highly competitive and demonstrated excellent classification potential.
Keywords: fuzzy logic; fuzzy inference systems; learning algorithms; hybrid algorithms; algorithms comparison.
International Journal of Business Intelligence and Data Mining, 2018 Vol.13 No.1/2/3, pp.267 - 290
Available online: 03 Nov 2017 *Full-text access for editors Access for subscribers Free access Comment on this article