In my data, I have a column with negative and positive values. Here negative value means how much some things are missing and positive values means unexpected additional things and 0 means neutral which is always expected value. So how can I use this column of data as input for deep belief networks. Can I input negative numbers as input in deep belief networks?
I think you can consider two things. Firstly, there should not be an underlying problem inputting negative values - is there are specific reason you question this?
More importantly if you want, you can pre-process your dataset, or do it during input. There are many activation functions you can use that result in absolute values, like something as simple as the sigmoid function. There is nothing wrong with activating input values, in fact its recommended.