I have modeled an ANN classifier that takes as input a sky image (extract 72 statistical features of RGB channels) and outputs a single class (is it a clear sky '1' or not '0').
Although the target is a column vector of 1s and 0s depending of what type of observation it is (clear or not), I have noticed that the output of prediction, after the training, is a number between 0 and 1 and not 0 or 1. This means that the classifier was not able to distinguish between a clear sky and a not clear one (which is the sole purpose of this work) and gave an approximation between the two (kind of)..
So my question is, do you think that using another output transfer function (other than logsig) would solve the problem? because I don't want to specify a manual threshold to differentiate between the classes (why use ANN then..). Thanks for your time and consideration.