Learning from DNA Microarray Data Using Nearest Neighbor Classifiers
Author Information
Author(s): Daniel Berrar, Ian Bradbury, Werner Dubitzky
Primary Institution: School of Biomedical Sciences, University of Ulster at Coleraine
Hypothesis
Is there a significant difference in performance between nearest neighbor approaches and more complex multiclass methods for DNA microarray data classification?
Conclusion
The k-NN classifier is a simple and effective alternative to more complex models for multiclass microarray data sets.
Supporting Evidence
- k-NN classifiers performed as well as or better than complex models.
- The study used a statistical significance test to assess performance differences.
- The k-NN model provides confidence scores for predictions.
Takeaway
This study shows that a simple method called k-NN can do just as well, or even better, than more complicated methods when classifying DNA data.
Methodology
The study applied a distance-weighted k-NN classifier to three gene expression data sets and compared its performance with other classifiers using statistical significance tests.
Potential Biases
Potential bias from the selection of data sets and the methods used for feature selection.
Limitations
The study may not generalize to all types of high-dimensional biological data beyond microarray data.
Participant Demographics
The study involved gene expression profiles from 60 human cancer cell lines of various origins.
Statistical Information
P-Value
0.38
Confidence Interval
72.10 ± 7.07% for NCI60 data set
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website