You are here:
A Balanced Accuracy Fitness Function Leads to Robust Analysis Using Grammatical Evolution Neural Networks in the Case of Class Imbalance
Citation:
Hardison, N. E., T. J. Faneli, S. M. Dudek, D. REIF, M. D. Ritchie, AND A. A. Motsinger-Reif. A Balanced Accuracy Fitness Function Leads to Robust Analysis Using Grammatical Evolution Neural Networks in the Case of Class Imbalance. Presented at Genetic and Evolutionary Computation Conference (GECCO), Atlanta, GA, July 12 - 16, 2008.
Impact/Purpose:
In the current study, we test the power of GENN to detect gene-gene interactions in data with a range of class imbalance using two different fitness functions (classification error and balanced error), as well as re-sampling. We show that when using the original fitness function, classification error, class imbalance greatly decreases the power of GENN. Using the novel fitness function, balanced error, GENN is robust to class imbalance. The re-sampling methods demonstrated some success, but not as promising as balanced accuracy. Based on the results of this study, balanced error has replaced classification error in the GENN algorithm.
Description:
The identification and characterization of genetic and environmental factors that predict common, complex disease is a major goal of human genetics. The ubiquitous nature of epistatic interaction in the underlying genetic etiology of such disease presents a difficult analytical challenge. This challenge has prompted the development of novel computational approaches, including Grammatical Evolution Neural Networks (GENN). GENN has been a highly successful endeavor, but has so far only been tested in simulations with balanced numbers of cases and controls. Real data, however, rarely has such perfectly balanced classes.