Average Gain Ratio and Correlation-Based Feature Selection for Imprecise Classification in Algorithm C4.5

The attribute split in the C4.5 algorithm has proven successful in building a classification in the form of a tree to facilitate understanding and interpretation. However, the attribute split process tends to choose attributes with high values, even though they do not necessarily play a major role in the classification results, and does not consider the correlation of attributes with labels, affecting classification performance. Average Gain Ratio (AGR) has been proven to overcome weaknesses and problems in the split process. Correlation-Based Feature Selection (CBFS) is also used in the split process to calculate the correlation of attributes with labels. This study uses the AGR method to overcome the problem of attribute split criteria and the CBFS method to select attributes that correlate with labels to increase the performance of the C4.5 algorithm. The process of selecting the split attributes using AGR and the comparison with the CBFS method was shown to improve the performance of the C4.5 classifier, as indicated by the average results for accuracy (87%), sensitivity (91%), G-mean (85%), AUC (84%), and cost (3.67) on six UCI datasets.