Just a couple of thoughts.

This is hard, I suggest, because (i) attractiveness is strongly culturally and fashion determined, and (ii) it is often determined by the way the features move rather than by their static configuration.

Asymmetry and deformity is always unattractive?

Not obvious to me that you need an intelligent system to assess attractiveness. E.G. sexual attractiveness has not got too much to do with intelligence?

Can you give some examples of non-geometrical facial features?  I don't understand what these would be.

You might use Scale & pose invariant features using Adaboost Machine learning feature extraction algorithm.

Haar features can be used

LBP might be used too.

Read on the "beauty mask" based on golden ratio (Marquardt ). 

The feature extraction method would be anyway geometrical. 

james sir, i think there is a misunderstanding. I was not able to convey my problem to you properly. What i am trying to do is to measure the attractiveness of an individual based on the structure of your face(not body). you can see a similar kind of effort in the attached link (http://www.anaface.com/ ). By intelligent system i mean a computer system based on the concepts of machine learning and artificial intelligence to determine the attractiveness score( nothing to do with an individual's IQ).

Adam Hughes, by non geometrical features i mean the one not derived from the ratio's of length/Area of face to nose,eyes etc., It may be texture, may be color,. One can use local descriptor as GLOH, SIFT to extract features from eyes, lips etc.. Actually i am looking for those and haven't found anything substantial. I can keep you posted if you wish.

Ashish Gor, I don't think pose and scale invariant features would help here as, i am initially working of the basic model and the faces i would consider should be of fixed scale and orientation. Regarding LBP i don't think would be helpful either as much i know it is more apt for classification related problem. can you please elaborate the significance of  using haar coefficients as features here as i havn't found any proper explanation.

Ransalu Senanayak, actually i have taken such ratios, but to make my system more efficient i am looking for some for different features.

http://www.anaface.com/

Sorry, not heard of other mathematical analysis of facial beauty. Maybe making use of Galois theory & Reimann surface help to efficiently work with 2D manifolds. 

Symmetry is often an indicator for attractiveness, but also it is the similarity to the person who judges the attractiveness, would be an interesting study to compare facial features from the one who rates the attractiveness and the subject being rated.Then compare familiarity of features between them with the rating given. Sorry that might not help you really with your question.