You are constructing a function from several inputs to meet a desired output. This function should factor the amount of each attribute in conjunction with the importance. Your use of value pairs for such a function also coheres excellently with a documented decision model. I believe that a weighted sum is in order.
The weighted sum model takes in each attribute's value and their weight and proceeds by multiplying the former by the latter. If an attribute carries great importance in the hit chance, then the weight should reflect this by carrying a higher value as well. For example, the AttackSkill attribute seems an important determinant, so it will be assigned a weight of 1.5. Conversely, AttackBonus impacts the positive chances much less, and is assigned 0.8. We are left with.
(AttackSkill*1.5)+(AttackBonus*0.8) = ATTACKPOWER
The same will be done on the defender's behalf accounting for an additional attribute.
(DefenseSkill*1.5)+(DefenseFitness*2.0)+(DefenseBonus*0.8) = DEFENSEPOWER
Weighted sum equations carry the advantage of modularity, allowing them to be scaled to any extent without breaking down. After the sums are obtained, they need to be referenced with each other to map these "forces" to a 0-10 chance. Here's a simple method.
We first measure the difference between the power of the attack and defense. If positive, the attack has 50% or greater of following through and vice versa. The difference is passed to a sigmoid function which presents a rather convenient chance curve.
The curve is mapped from 0-10 and voila.
(sigmoid(ATTACKPOWER-DEFENSEPOWER)+1.0)*5.0 = HITCHANCE