Özet:

Due to the diversity of sources, a large amount of data is being produced and also has various problems including mislabeled data, missing values, imbalanced class labels, noise and high dimensionality. In this research article, we proposed a novel framework to address high dimensionality issue with feature reduction to increase the classification performance of various lazy learners, rule-based induction, bayes, and tree-based models. In this research, we proposed robust Quarter Feature Selection (QFS) framework based on Symmetrical Uncertainty Attribute Evaluator. Our proposed technique analyzed with Six real world datasets. The proposed framework , divide whole data space into 4 sets (Quarters) of features without duplication. Each such quarter has less than or equals 25 % features of whole data space. Practical results recorded that, one of the quarter, sometimes more than one quarter recording improved accuracy than the already available feature selection methods in the literature. In this research, we used filter-based feature selection methods such as GRAE, IG, CHI-SQUARE (CHI 2), Relief to compare the quarter of features produced by proposed technique.

Anahtar Kelimeler: