GRASP for Instance Selection in Medical Data Sets

Abstract Medical data sets consist of a huge amount of data organized in instances, where each one contains several attributes. The quality of the models obtained from a database strongly depends on the information previously stored on it. For this reason, these data sets must be preprocessed in order to have fairly information about patients. Data sets are preprocessed reducing the amount of data. For this task, we propose a GRASP algorithm with two different improvement strategies based on Tabu Search and Variable Neighborhood Search. Our procedure is able to widely reduce the original data keeping the most relevant information. Experimental results show how our GRASP is able to outperform the state of the art methods.

• Type: Journal
• Year: 2010
• Where: ISAMI
• Authors: Alfonso Fernández, Abraham Duarte, Rosa Hernández, Ángel Sánchez
• Keywords: Abstract Medical Data, Data Sets, Emerging Technology, ISAMI 2010, Variable Neighborhood Search