What is precision on prediction given precision of rules I'm new to statistics and have a question about my machine learning method. This method can distinguish between positive/negative examples. The result is a set of rules to define only positive examples.
For each rule I have calculated the precision as follows:
precision = $\frac{\mathrm{TP}}{\mathrm{TP+FP}}$
where:


*

*$\mathrm{TP}$ = true positives covered by the rule.

*$\mathrm{FP}$ = False positive covered by the rule.


I use the method to predict an unknown example. The example is predicted as deleterious by three independent rules, which have precisions of


*

*30% (a bad rule)

*95%

*88%


What is the final precision score for the unknown example?
 A: (This could have been a comment but gets a bit too long).
The "final precision score" seems vaguely defined here.  If you just had three independent experiments trying to measure the same underlying parameter (let's call it "truthness"), you could just pool your results and work out the overall score of $\frac{\sum{TP}}{\sum{TP + FP}}$.  But in fact it looks like you want to compare three independent experiments that are measuring different underlying parameters - the truthness as measured by rule 1, by rule 2 and rule 3.  There's no statistical way of judging between the three rules - perhaps rule 1 is the best measure and it's a really bad machine learning method, perhaps rule 3 and 2 are and it's good.  You need some extra (non-statistical) information to judge which of the rules is more useful for you.  
A: I think your question is not very well asked : when you say "What is the final precision score for the unknown example", I guess you mean what is the final precision score for the method because you cannot calculate precision for one individual (see your formula in your question). 
And if I understand correctly you thus need to define your new method using the three rules. How do you aggregate them to make a prediction ? Majority vote ? When you have defined how to aggregate the three rule, you can use your prediction to recalculate your precision the same way you did for the three rules. 
