To aggregate or not to aggregate high-dimensional classifiers

Abstract

BackgroundHigh-throughput functional genomics technologies generate large amount of data with hundreds or thousands of measurements per sample. The number of sample is usually much smaller in the order of ten or hundred. This poses statistical challenges and calls for appropriate solutions for the analysis of this kind of data.ResultsPrincipal component discriminant analysis (PCDA), an adaptation of classical linear discriminant analysis (LDA) for high-dimensional data, has been selected as an example of a base learner. The multiple versions of PCDA models from repeated double cross-validation were aggregated, and the final classification was performed by majority voting. The performance of this approach was evaluated by simulation, genomics, proteomics and metabolomics data sets.ConclusionsThe aggregating PCDA learner can improve the prediction performance, provide more stable result, and help to know the variability of the models. The disadvantage and limitations of aggregating were also discussed.

Gene Expression Profiles for Predicting Metastasis in Breast Cancer: A Cross-Study Comparison of Classification Methods

“…Our finding of voting superiority agrees with four other studies: one using multiple different feature extraction methods in combination with SVM for gene microarray classification showed that using a voting-based method across all the examined combinations achieved a better 10-fold cross-validated classification performance compared to any single combination [33]; a second study showed that an SVM-based ensemble outperformed single SVM for microarray data classification [42]; a third study comparing the performance of principal component discriminant classifiers either with or without voting using cross-validation applied on a simulated dataset a leukemia microarray gene expression dataset, a Gaucher serum proteomics dataset and a grape extract metabolomics dataset, also showed that voting had a better performance than the nonvoting method [43]; a fourth study comparing the performance of single models to combined models in thirteen diverse microarray datasets, which included predictions of estrogen receptor positivity and complete pathological response to chemotherapy in breast cancer, found that the majority of combined multiple models had a better classification performance than the single models [21]. Furthermore, our findings also agree with a study by Taylor and Kim who, by splitting their original datasets into training and test parts, showed that voting based on nearest mean voters was a top performing method with respect to classification on lung or prostate cancer data.…”