compare int arrays for 'similarity' - more accurate than weighted average?

say there is a number of arrays with length 12, containing signed integers in a range of roughly ±100, how can i compare the 'signature' or 'harmonic content' of these arrays to each other, in a way that is more accurate than a simple weighted average?

Would i 开发者_开发知识库have to look into neural networks (if this even would be suitable, i don't know tbh) or is there a more straightforward way?

EDIT:

on a graph, imagine the index on the x-axis and the value on the y-axis. two arrays would be very similar if the shape of the graphs would be similar after normalization.

i find it hard to find the right words for this problem.

The arrays could be treated as signals, hence you should try to do a scaled cross correlation between the two signals, or extract frequency domain info and compare their spectrum ( FFT of both signals ).