Training data selection method for prediction of anticancer drug effects using a genetic algorithm with local search

Here, we propose a training data selection method using a Support Vector Machine (SVM) to predict the effects of anticancer drugs. Conventionally, SVM is used for distinguishing between several types of data. However, in the method proposed here, the SVM is used to distinguish areas with only one or two types of data. The proposed method treats training data selection as an optimization problem and involves application of a genetic algorithm (GA). Moreover, GA with local search was applied to find the solution as the target problem was difficult to find. The composition method of GA for proposed method was examined. To determine its effectiveness, the proposed method was applied to an artificial anticancer drug data set. The verification results showed that the proposed method can be used to create a verifiable and predictable discriminant function by training data selection.