0
$\begingroup$

First of all, I just entered the topic of artificial neural networks, I'm sorry if my question sounds ridiculous.

I want to estimate social return to education for Argentina using neural networks. But there's not sufficient data set for Argentina. So I want to train my neural networks with another country's data set, for example Brazil. If that's possible, my estimation for Argentina will be consistent? So I'm curious whether country-specific data sets important for neural network analysis. If I trainneural networks with the data of a country, can I get generalizable results when I make predictions for a different country?

Thanks in advance.

$\endgroup$
1
  • $\begingroup$ If Argentina is exactly like Brazil then you can generalise from one country to another. If they are completely different then no you cannot. This is not an issue restricted to neural networks it applies to any technique. $\endgroup$ – mdewey Aug 25 '20 at 14:37
0
$\begingroup$

You could do transfer learning, i.e., train a neural network (NN) architecture on dataset $A$, and then use a similar NN architecture to continue learning on dataset $B$. For instance, you may want to remove the input and final nodes of the NN trained on $A$ and replace them with new (untrained) units and weights to continue training on $B$.

Note, however, that if you don't have a huge dataset, an NN may not be your best bet. You may want to consider simpler models at first such as a Linear Regression, a regression SVM, Decision Tree, Random Forests, etc.

$\endgroup$

Not the answer you're looking for? Browse other questions tagged or ask your own question.