Keeping things in short. I want to distinguish two classes on a feature space. All what I know are:
- the theoretical distributions of the class 0 data on each feature dimension.
- the expected distance between the mean of the class 1 data and the mean of the class 0 data on each dimension.
but I do not have any labeled data. I wonder whether LDA can help in this case and how to do it.
Previously, I tried hypothesis testing, but it is very difficult to decide the significance level on each dimension: when I plotted the distribution of accepted data, their sample distributions are far away from the theoretical, often due to the fact that the variance of each class is of a similar scale of the difference between the means of two classes.
PS: all features are correlated.