I have a set of products P {1...n} which are rated on a goodness scale G ={1...100} (G10 is more good than G5). Each product has a set of features F {1....m}, now I want to learn a model for goodness, expressed as a combination of features values. Something like (F1 > 100) AND (10< F2<20) AND (F3=blue) is Good with a confidence of 40%. etc.
What would be the best learning algorithm to use?
My analysis is :
1) Association rules: Add goodness as a feature of the product and repeat the product in the set equal to the number of goodness rating for it. So a G5 product will be repeated 5 times, so that it gets 5X weight. But the problem I see is if the final rules are like (F1 < 100) = (90% confidence), (F3 = blue) = (85% confidence) etc and a compound rule with F1 AND F2 AND F3 = (20% confidence) which rules would I select? Also if all the products had the same value for one feature Fx then it would be the one with 100% confidence, which does not give me any good info.
2) Decision tree classifier: Divide the products into 2-classes of Good and Not-Good w.r.t to the average Goodness of the set. So all products below average goodness are in class NotGood. Also repeat the products in their class depending on the value of goodness, ex if AvgG= 20, a product with G18 is in class NotGood and is repeated (20 - 18 =) 2 times and a product G10 product is repeated (20-10=) 10 times, and a product with G40 is repeated (40-20=) 20 times (here the class is Good). This might take care of the case when all the products have same value for a feature as well.
Please suggest which approach seems better and if there is any other alternate approach. Also please provide some links if someone has already worked on a similar problem.
