Bio-inspired machine learning in microarray gene selection and cancer classification

Microarray technology today has the ability of having the whole genome spotted on a single chip. It allows the biologist to inspect thousands of gene activities simultaneously. Machine learning approaches are suited and used to discovering the complex relationships between genes under controlled experimental conditions and classify microarray data by identifying a subset of informative genes embedded in a large data set that involves multiple classes and is infected with the high dimensionality noise. In this paper, a hybrid system integrates genetic algorithms and decision tree is proposed for genes expression analysis and prediction to their functionality for cancer classification. The learning capacity of decision trees used in the base learning systems is boosted by feature selection method. Experiments present preliminary results to demonstrate the capability of hybrid system to mine accurate classification rules for classifying prediction in comparable to traditional machine learning algorithms.