# Neural network architecture for formation classification based on 2d spatial data

I want to build a neural network to classify different soccer team formations like 5-3-2 or 4-3-3 (defense, midfield, attacker).

For this purpose I generated labelled 2d spatial data representing each of the 10 players on the field (no goalkeeper). The pitch has the following dimension: [-55, 55, -30, 30].

One datapoint may look like this:

[[-17.07, 23.79], [-29.4, 12.59], [-28.18, 9.42], [-28.88, -3.23], [-23.65, -19.42], [-12.19, 12.64], [-12.94, 0.01], [-12.12, -8.61], [1.88, 6.42], [3.03, -4.17], [1.0]]


Ten coordinates of players on a pitch and one label (last item) representing the formation.

Which neural network architecture could work well with this type of data?

I'm thinking about trying a convolutional network but I'm not sure if there is a more suitable solution for this problem.