Sacrificing Overall Classification Quality to Improve Classification Accuracy of Well-Sought Classes
Kevin Michael Amaral, Ping Chen, Wei Ding, Rajani Shankar Sadasivam · 2016
Classification has been an active field in machine learning for decades. With many methods proposed for various topics in classification, this paper intends to show some initial ideas and findings in one classification scenario where accuracy of only one or a few classes is greatly valued, while the other classes are not important. Using a neural network model and challenging real world dataset, our preliminary results showed the accuracy of important class was significantly improved by sacrificing the accuracy of unimportant classes.