Which machine learning approach should I use to estimate how many products a research group should have to improve its category?

Question

Currently, in my country, there is a system in which certain groups of researchers upload information on products of scientific interest, such as research articles, books, patents, software, among others. Depending on the number of products, the system assigns a classification to each group, which can be A1, A, B and C, where A1 is the highest classification and C is the minimum. According to the classification of the groups, they can compete to receive monetary incentives to make their research.

At the moment, I am working on an application that takes the data of the system that I mentioned previously. I am able to say what classification the group currently has because we develop a scraper that counts the products and there is another service that is in charge of implementing all the mathematical model that the system has to calculate the category of the group.

But what I want to achieve is that my application would be able to give an estimate of how many products a research group should have to improve its category. I want to know if I can do that using neural networks.

For example, if there is a category C group, I want the application to tell the user how many articles and books it would make its category go up to B.

From what I have seen in some web resources, I could insert a training set into the neural network and have it learn to classify the groups, but I think it is unnecessary, because I can do that mathematically.

But I do not understand if it is possible for a neural network to process the current category that the group has and be able to give suggestions of how many products it needs to improve its category.

I think it must be a neural network with several outputs, so that in each one it throws the total for each one of the products, although it is not necessary to list all the products that the measurement model contemplates. But it is necessary for the network to learn which products are handled by a certain group, for example if a group does not write books, avoid suggestions that contemplate the production of books for the improvement of the category that the group has.

Answer 1

I believe you want a neural network that can predict future values of multiple variables given multiple inputs. This belongs to the general time series forecasting problem.

One of the best neural network architectures that can handle this problem is the LSTM, which is a type of Recurrent Neural Network. Their architecture allows them to develop a memory of what they have seen in the past and use it for future predictions. In other words, they can cross-correlate in a linear/nonlinear fashion several past steps of multiple variables to future values of multiple other variables, like a black box.

A useful tutorial for your purposes is this.