Evolutionary Search of Thresholds for Robust Feature Set Selection: Application to the Analysis of Microarray Data
We deal with two important problems in pattern recognition that arise in the analysis of large datasets. While most feature subset selection methods use statistical techniques to preprocess the labeled datasets, these methods are generally not linked with the combinatorial properties of the final solutions. We prove that it is NP-hard to obtain an appropriate set of thresholds that will transform a given dataset into a binary instance of a robust feature subset selection problem. We address this problem using an evolutionary algorithm that learns the appropriate value of the thresholds. The empirical evaluation shows that robust subset of genes can be obtained. This evaluation is done using real data corresponding to the gene expression of lymphomas.
Unable to display preview. Download preview PDF.
- 1.Davies, S., Russell, S.: NP-completeness of searches for smallest possible feature sets. In: Greiner, R., Subramanian, D. (eds.) AAAI Symposium on Intelligent Relevance, New Orleans, pp. 41–43. AAAI Press, Menlo Park (1994)Google Scholar
- 7.Weihe, K.: Covering trains by stations or the power of data reduction. In: Battiti, R., Bertossi, A. (eds.) Proceedings of Algorithms and Experiments (Alex 98), Trento, Italy, pp. 1–8 (1998)Google Scholar