Classifier ensembles have produced promising results, improving accuracy, confidence and most importantly feature space coverage in many practical applications. The recent trend is to move from heuristic combinations of classifiers to more statistically sound integrated schemes to produce quantifiable results as far as error bounds and overall generalization capability are concerned. In this study, we are evaluating the use of an ensemble of 8 classifiers based on 15 different fusion strategies on two medical problems. We measure the base classifiers correlation using 11 commonly accepted metrics and provide the grounds for choosing an improved hyper-classifier.