Loading [a11y]/accessibility-menu.js
Naïve Bayes ensemble: A new approach to classifying unlabeled multi-class asthma subjects | IEEE Conference Publication | IEEE Xplore

Naïve Bayes ensemble: A new approach to classifying unlabeled multi-class asthma subjects


Abstract:

Heterogeneity of complex human diseases makes it difficult to control and treat diseases effectively. Recent clinical efforts involve redefining subtypes of the diseases ...Show More

Abstract:

Heterogeneity of complex human diseases makes it difficult to control and treat diseases effectively. Recent clinical efforts involve redefining subtypes of the diseases using cluster analysis. However, follow-up studies for the new clusters/subtypes identified from these efforts have been limited by the fact that clinical datasets collected from different times and cohorts often have different subjects and variables. A typical way to address this problem is to use only the common subjects and the variables among the datasets for analysis, which will inevitably result in loss of power and information. In this work, we developed a supervised learning strategy and a new classification algorithm called Naïve Bayes ensemble, which can classify unlabeled subjects from a new dataset into the clusters/subtypes previously identified from an earlier dataset so that one can make full use of the new dataset for further analysis. The Naïve Bayes ensemble employs a probability framework to summarize votes from single classifiers and take into account misclassification rates of each class predicted by each single classifier. We also propose a procedure to automatically select single classifiers to be included in the ensemble, which are based on both classification performance and the diversity of the single classifiers. The Naïve Bayes ensemble outperforms 7 single classifiers and other ensemble methods when we evaluated their performance using leave-one-out cross validation (LOOCV) on our asthma clinical datasets. Furthermore, using biological functional validation, our results show that the supervised learning strategy and the Naïve Bayes ensemble we proposed can indeed greatly increase the power of the downstream analysis and help yield biologically/clinically meaningful results.
Date of Conference: 15-18 December 2016
Date Added to IEEE Xplore: 19 January 2017
ISBN Information:
Conference Location: Shenzhen, China

References

References is not available for this document.