Analysis of the Effect of Feature Reduction on Accuracy and Computational Time in Mushroom Dataset Classification

Abstract

Classification is a technique to mapping the class of a certain data from its attribute or feature values. One of things that affects the classification result is the correlation of its features to the class classification results. Research conducted to determine the effect of the reduction in features that are least correlated or have a distant relationship with the classification result class (dependent variable). Because features that do not have much correlation, have no effect on the classification results. From the research, the accuracy of the reduction of each feature per test scenario has a range between 83% -88% higher than the initial accuracy without feature selection at 82% accuracy. Meanwhile, the computation time obtained does not have a significant difference in changing compared to without feature reduction, in the range of 2.3-2.7. For the data used is the Mushroom dataset obtained from the UCI Machine Learning Repository