TY - EJOU
AU - Ezhilarasi, L.
AU - Shanthi, A.P.
AU - Maheswari, V. Uma
TI - Rough Set Based Rule Approximation and Application on Uncertain Datasets
T2 - Intelligent Automation \& Soft Computing
PY - 2020
VL - 26
IS - 3
SN - 2326-005X
AB - Development of new Artificial Intelligence related data analy sis methodologies w ith rev olutionary
information technology has made a radical change in prediction, forecasting, and decision making for
real-w orld data. The challenge arises w hen the real w orld dataset consisting of v oluminous data is
uncertain. The rough set is a mathematical formalism that has emerged significantly for uncertain
datasets. It represents the know ledge of the datasets as decision rules. It does not need any metadata.
The rules are used to predict or classify unseen ex amples. The objectiv e of this research is to dev elop
a rough set based classification sy stem that predicts and classifies unseen ex amples by learning from
the minimal subset of decision rules ex tracted from uncertain datasets using rule approx imation. This
paper proposes a nov el rule approx imation classifier, Weighted-Attribute Significance Rule
Approx imation (WASRA) that uses a subset of the decision rules generated by any rule induction
algorithm, to compute the concept w eights of the condition attributes. The concept w eights and the
significance of condition attributes are used to design a nov el classifier. This classifier is implemented
and initially tested on a few benchmarked datasets of the UCI repository . The classifier is subsequently
tested on a real-time dataset and compared to other standard classifiers. The ex perimental results
illustrate that the proposed WASRA performs w ell and show s an improv ement in the prediction
accuracy compared to other classifiers. This classifier can be applied to any dataset w hich has
uncertainty .
KW - Rough set theory
KW - concept weight
KW - rule approximation
KW - attribute significance
DO - 10.32604/iasc.2020.013923