Enhancing the Quality of Noisy Training Data Using a Genetic Algorithm and Prototype Selection.
Boseon Byeon, Khaled Rasheed, Prashant Doshi · International Conference on Artificial Intelligence · 2008
This paper introduces a novel technique to enhance the quality of training data with noisy dependent variable for binary classification. Noise reduces classification accuracy by disrupting the training data set and causing the classifier to build incorrect models. Our approach (GAPS) uses a genetic algorithm (GA) to create the set of suspicious noisy instances and prototype selection (PS) to identify the set of actual noisy instances. This paper shows that the combination of genetic algorithm with prototype selection enhances the quality of noisy training data sets and increases classification accuracy of the model built with the enhanced training data.