Combined Kernel Function Approach in SVM for Diagnosis of Cancer

A Study of Crossover Operators for Gene Selection of Microarray Data

“…The table gives the number of genes and the classification accuracy reported by each author (Reported ) and the classification accuracy obtained by our approach (GeSeX) when we fix the number of genes to the value used in the corresponding paper. [17] [ our method, that means that for each data set and each previously cited work [17,22,8,16], we determine which classification accuracy can be obtained by our GA for the number of genes reported in this work. Moreover, we evaluate the classifier accuracy with the same number of runs: for [17] and [22], the result is the best accuracy obtained in one run while for [8] and [16], this is the average over 10 runs.…”

“…[17] [ our method, that means that for each data set and each previously cited work [17,22,8,16], we determine which classification accuracy can be obtained by our GA for the number of genes reported in this work. Moreover, we evaluate the classifier accuracy with the same number of runs: for [17] and [22], the result is the best accuracy obtained in one run while for [8] and [16], this is the average over 10 runs. We also use the same test samples as the authors for each dataset, this is important because previous studies have shown that the accuracy estimate may be biased and may have an important variance [3].…”

“…The first compares the crossover operator GeSeX with two well-known crossover operators. In the second study, we carry out a comparison with four highly effective GA-based gene selection approaches [17,22,8,16]. …”

“…In [16] the authors combine SVM and GA in another way. Their SVM uses a kernel function that combines a set of simple kernel functions and they propose a new learning method exploiting Evolutionary Algorithm technique to obtain an optimal decision model.…”

“…The literature offers a large number of solution methods for gene selection which are based on genetic algorithms, often combined with other approaches [19,6,18,17,8,16,4,22]. For instance, the so-called wrapper approach uses GAs to search over the space of gene subsets, the fitness of each subset being evaluated by its classification performance obtained by a given classifier.…”