Hey Kirstie, 

This sounds like a classic machine learning problem.  Is the input data labeled?  If so, I recommend checking out The Elements of Statistical Learning by Hastie et al.  It provides a nice overview of what can be learned by competing classification techniques.  If the data is not labeled, my best reference is Deep Learning by Goodfellow.  However, I must provide the disclaimer that it is not a book that I adore.

Do you have a new algorithm for pattern detection?  Or have you discovered an empirical property within a class of data that is best predicted by a particular mix of algorithms?

Oh, I will make one more recommendation.  Pedro Domingo's The Master Algorithm.  

See you at ICAC!

There is a good overveiw of metrics and their properties in [1] Sokolova, Marina ; Lapalme, Guy: A systematic analysis of performance measures for classification tasks. In: Information Processing and Management 45 (2009), Nr. 4, P. 427–437 and in [2] Ferri, C. ; Hernández-Orallo, J. ; Modroiu, R.: An experimental comparison of performance measures for classification. In: Pattern Recognition Letters 30 (2009), Nr. 1, S. 27–38. URL http://www.sciencedirect.com/science/article/pii/S0167865508002687.

But I fear this is not what you asked for, because these are "classic" performance measures for classification or prediction. If I understand you correct. You are looking for the overall capability to discover relevant pattern. Since the pattern are in a broader view statistical how would you measure thier relevance without problem or even domain knowledge? Further, how can you evaluate the "finding" ability of an algorithm in gerneral? The closest I assume is something like gini-index.

Hi Thomas and Chris, First, thank you both for your responses.  Thomas is on the right track when he says that I am looking for more than comparing the performance of learning algorithms per se. Rather I am looking at the ability of an algorithm to discover relationships/patterns of interest. Of course, the most work on such discovery or 'finding ability' is on the comparison of  learning algorithms or pattern recognition algorithms.  Thomas, I found your comment on the gini-index intriguing and I'll follow that path. I am excited about some of this new work - the purpose of the question as I stated was to see if I could add to the current literature review in an upcoming publication.  When that is done, I will gladly send you a copy and we can discuss in depth the approach I have taken.