Improved Classification Accuracy by Feature Selection using Adaptive Support Method

Abstract

The explosion of data which is happening now must be utilized to support decision making both in terms of business and other matters. Data which are becoming assets today needs to be analyzed and extracted in order to find valuable information. The results of data analysis can be used to make predictions, one of which is classification. For high dimensions data, we require preprocessing stage so that the model building process is not complex and the analysis is accurate. One of the preprocessing stages that need attention is feature selection. Feature selection is applied to reduce features without diminishing the accuracy and information in the data. Performing feature selection can also be done by using the association rule. Association rule refers to considering the association relationship between items and the frequency of items occurrence as features. However, the obstacle in implementing the association rule is when determining the minimum support value. Therefore, an adaptive support method is proposed to determine the minimum support value automatically based on the characteristics of the dataset. In this present study, a feature selection method using adaptive support is proposed. Based on the experimental results using 3 classifiers, the accuracy and F1-Score values for the feature selection method using adaptive support are higher compared to the Information gain method.