Ant colony optimization and crazy particle swarm optimization for support vector support machine classification on high-dimensional dataset

The data generated by DNA microarray technology can be used to predict and classify genes taken from certain tissues in humans to be classified as cancer or not. Microarray data consists of thousands of variables, but limited data is available. Support Vector Machine (SVM) is a supervised learning method that can be used for classification on the high-dimensional dataset. There are two problems in SVM classifier that influence the classification accuracy, which are tuning SVM parameters and selecting the best features subset to the SVM classifier. Several approaches have been carried out for the feature selection process and tuning SVM parameter, including a wrapper-based approach. The wrapper-based algorithm used in this research is Crazy Particle Swarm Optimization (CRAZYPSO) and Ant Colony Optimization (ACO). Both algorithms are the computational intelligence-based algorithm that can be used to solve the optimization problems, such as feature selection and parameter optimization. These algorithms are inspired by animal behavior in the real world. CRAZYPSO calculations are very simple compared to other optimization algorithms. While ACO has several advantages, such as strong robustness, well-distributed computing mechanism and easily combined with other methods. This study wants to compare the CRAZYPSO and ACO algorithm in the case of microarray data classification. The microarray datasets used in this study are the prostate dataset and colon dataset. This study uses k-fold cross-validation accuracy to compare the CRAZYPSO and ACO algorithm in the case of microarray data classification using Support Vector Machine. The result shows that the ACO algorithm gives a better result in feature selection than the CRAZYPSO algorithm with higher accuracy rate and less selected features. This study also shows that the SVM parameter optimized using ACO algorithm gives higher classification accuracy rate than parameter optimized using CRAZYPSO algorithm.