Can association mining deal with non-linear relationships?

Question

For an association rule x->y, When the lift is 1, x and y are independent. If lift is lesser than 1, then both are negatively correlated else they are positively corrrelated. What if they have non-linear correlation? Would the association rule/frequent itemsets capture such associations?

Answer 1

Association rule mining works with discrete items (nominal labels) and is based on contingency tables (counts). An appropriate type of correlation is the Phi correlation coefficient. The concept of non-linear relationships does not apply in this case.