Factors affecting boosting ensemble performance on DNA microarray data

Boosting techniques have been applied to DNA microarray data because their high dimensionality has made them difficult to analyze. However, classification performance varies between boosting algorithms. We have investigated factors affecting error in boosting ensemble classifiers on DNA Microarray data: number of training samples, number of boosting iterations, complexity of base learners and diversity of models. Specifically we have applied diversity measures to investigate the relationships between model type, model accuracy, diversity and ensemble accuracy.