March 1, 2002Journal of the American Statistical Association
Comparison of Discrimination Methods for the Classification of Tumors Using Gene Expression Data
Authors
Discussion
Loading...
Member takes
Implication
Comparative benchmark study reveals differential performance across statistical and machine learning classifiers on tumor gene expression profiles, highlighting optimal algorithms for cancer...
Key Points
- To evaluate and compare the performance of various statistical discrimination methods and machine learning algorithms for classifying tumors using microarray gene expression profiles.
- Benchmarked classical discrimination methods, including nearest-neighbor classifiers, linear discriminant analysis (LDA), and classification trees.
- Evaluated advanced machine learning ensemble techniques, specifically bagging and boosting algorithms.
- Applied and compared all classification methods across datasets from three published cancer gene expression studies using cDNA microarrays and high-density oligonucleotide chips.
- Demonstrated that discrimination algorithms can successfully differentiate known and novel tumor classes using high-dimensional gene expression data.
- Identified distinct performance tradeoffs among nearest-neighbor classifiers, linear discriminant analysis, decision trees, and ensemble machine learning approaches across the three cancer datasets.
Cite This Study
Dudoit et al. (2002) studied this question.