An Improved Accuracy of Multiclass Random Forest Classifier with Continuous Attribute Transformation Using Random Percentile Generation

This study aims to improve classification accuracy by transforming continuous attributes into categories randomly generating percentile values as categorization limits. Four algorithms were compared for the generation of and selected based on small variability distribution highest revenue expectations. The testing training data becomes second consideration. Random forest (RF) is modeled from percentiles with three transformation variations. results ANOVA test, algorithm variations transformation, has a mean that not significantly different best model original dataset model. However, in some data, RF attribute was superior effectiveness this very well applied LR, MLP, NB methods. In tuition fee dataset, application methods each had an 0.178, 0.204, 0.318. give significant increase 0.967, 0.949, 0.594 method, respectively. date fruits effective MLP method 0.193 (original attribute) 0.690 (continuous transformation). are effectively MPL, datasets categorical mixed attributes.