Because this is a Anomaly Detection problem.
So when you do the classification, in the training stage, you only use one-class-labeled data? or use both classes of data for anomaly?
In addition to Sangeen’s answer, I would like to say that the factors influencing your decision are:
1- The nature of the problem (detection or classification),
2- The algorithm employed,
3- And the dataset used.
Firstly, you have to determine precisely whether you problem is detection or classification. Secondly, you need to select successfully the algorithm to be used, and finally, you have to search for dataset suits your work.
For anomaly detection problem, the best scenario is to use one-class training dataset elements that reflex the normal situation. This will help you establishing fixed boundaries between the normal and suspicious situations. In the testing, any unseen element with significant deviation from the normal space will be marked as suspicious.
All the best
Hi Zhiliu,
To performs one-class classification on a dataset, you ONLY need to have one target label.
HTH.
Samer
Anomaly Detection problems usually show class imbalance characteristics which means there will be significant difference between number of instances from both classes. If your dataset is of that kind, I would suggest undersampling of the normal instances or oversampling of the anomalous instances and a two class classification afterwards.However, if anomalous instances are very small in number or cannot be represented adequately,go for one-class classification.
All the best.
Dear Zhiliu
All classifications algorithm design for binary class classification problem then they extended to deal with multi-class classification problem. so, if your problem is binary class use binary other wise use multi class.
Regards,