Question

The general neural network methodology is often described as a "black box" methodology. In a practical sense, which of the following is the most negative effect of this?

A. The predictions from neural nets are often accurate, often more accurate than from other methods, but it is practically impossible to untangle the effects of the individual explanatory variables on the dependent variable.

B. The detailed steps of the algorithm are totally random, even though they often produce accurate predictions.

C. Even though the algorithms use linear functions of the explanatory variables, it is practically impossible to interpret the coefficients of these linear functions.

D. The algorithms can have hidden layers and hidden nodes, which make no theoretical or practical sense at all, even though they can result in accurate predictions.

Answer 1

Answer is A

A. The predictions from neural nets are often accurate, often more accurate than from other methods, but it is practically impossible to untangle the effects of the individual explanatory variables on the dependent variable.