I'm trying to generate prediction of the column "dubina" using this algorithm made in R's "neuralnet" package. But I keep getting non-reliable neural-net output. I have tried changing the number of hidden layers, normalizing and denormalizing data. Is there a mistake in the algorithm, maybe because of the activation function being logistic, not sigmoid? The algorithm and the dataset are added as attachments. I'll be very grateful if you'd help me. Marijan