Is there an LDA method that matches the format "D_t= W_Tc^T * E_t + w_0"

Where W_Tc^T are the weights, E_t the signal/data and w_0 the offset. With D_t < 0 means E_t is class 1, and D_t > 0 means E_t is class 2.

Trying to replicate some work on Linear Discriminant Analysis but the current matlab classifier gives a discriminant for each class, with the higest discriminant being the selected class. I need it formatted as a threshold. Thanks,

Answer 1

[Ilya]

Sounds like you want a linear regression model:

http://www.mathworks.com/help/stats/linear-regression-model-workflow.html

If you have an older version of MATLAB without the LinearModel class, you can use the regress function:

http://www.mathworks.com/help/stats/regress.html

If you use an LDA implementation from the Statistics Toolbox, it predicts class labels and class posterior probabilities. You can threshold on the predicted posterior probability. Of course, that threshold should not be zero. By default it is 0.5.