Knowledge acquisition from mass spectra of blood samples using fuzzy decision tree and genetic algorithm

Recently, pattern analysis of mass spectra of blood samples has attracted attention as a promising approach to early detection of cancer. However, many questions have been raised about the reliability of the reported results due to the ldquoblack boxrdquo methods employed. The main objective of this paper is to introduce a simple framework of rule building procedure which results limited number of significant linguistic rules using them the clinician can explore the knowledge hidden in raw mass spectra of blood samples and consequently evaluate the cancer status in a new sample, without dependency on complex ldquoblack boxrdquo processing. To achieve this goal, we utilized two major branches of computational intelligence: fuzzy systems and evolutionary computing. We applied fuzzy decision trees as a powerful tool of building efficient fuzzy rules and in parallel, utilized genetic algorithm to optimize the number of the rules. Finally, we compared the performance of the proposed method with two well-known classification methods: KNN and LDA, and the results show excellence of our algorithm.