Do Artificial Neural Networks really improve the accuracy of forecasting models?
I think a starting place might be to consider if natural neural networks, aka brains can improve the accuracy of forecasting models? I generally take the answer to both questions as sometimes! Neural networks includes a number of technologies and can be implemented in various sizes and depths. The networks also require either training or statistical techiques to set the weights and connectivity. Finally, just like any forcasting, prediction situation. the variability and variations in what you are modeling relates to how well you can predict. Such variability needs to be represented in the training set. Static networks, well trained, can also be made weaker if what needs to be predicted is changing over time, or is changed by the use of the predictions. All of these factors add up, taking us back to the limitations of natural brains, 🙂 capacity, training, the nature of the prediction problem. It is a minor point, but an area still under research is what can be done with multple netwo
