It looks like your network is just predicting all observations into class 5. If your test dataset is very unbalanced, so nearly everything is class 5, this could lead to OK test accuracy, but the model is functioning poorly. You may want to try some sort of alternative sampling of your unbalanced data, e.g. take multiple observations from classes 1-4 for every observation of class 5. See this MATLAB Answers post for a few links and ideas.
Neural Network Confusion Matrix
3 次查看(过去 30 天)
显示 更早的评论
Hi,
Confusion matrix looking bad in my process but it has good accuracy result. Error on test data is 0.171886. I tried to be careful for my data distribution is to be good. What do you think about the problem?
Thanks and Regards.
0 个评论
回答(1 个)
另请参阅
类别
在 Help Center 和 File Exchange 中查找有关 Deep Learning Toolbox 的更多信息
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!