The microarray dataset contains a series of samples with the number of variables that reach thousands of genes expression. DNA microarrays are used to determine the level of gene expression and gene sequence in the sample. In cancer research, microarrays are used to study variation of molecular between tumors in order to develop better diagnosis and treatment for this disease. Classification is one of the important methods in microarray research to classify gene expression. Some characteristics of microarray dataset are high dimensions and imbalanced. Those characteristics cause prediction of classification which is over fitting. The purpose of this study is to overcome that problem with selection variables and generate synthetic data. The method for variables selection is Ant Colony Optimization (ACO), this method will compare with Genetic Algorithm (GA). The ACO method was inspired by the behavior of ant colonies looking for the shortest distance between the nest and food sources. The Method to solve imbalanced data is Synthetic Minority Oversampling Technique (SMOTE). This method generates synthetic data in minor classes randomly. In this study, the Support Vector Machine (SVM) is used to classify microarray dataset. This study uses breast cancer and lymphoma dataset. These datasets have different imbalanced ratios and number of variables. The result is variable selection using ACO method has fewer variables selected and higher AUC than GA method, but GA method more efficient in running time. SVM with SMOTE has higher performance than SVM without SMOTE.