Hyper-Parameter Optimization in Support Vector Machine on Unbalanced Datasets Using Genetic Algorithms

Hyper-parameter optimization and class imbalance are two challenging problems for machine learning in many real-world applications. A hyper-parameter is a parameter whose value is used to control the learning process and it has to be tuned in order to reach good performance. The class imbalance occurs when one class contains significantly fewer instances than the other class. Common approaches for dealing with the class imbalance problems involve modifying the data distribution or modifying the classifier. This paper presents an optimization framework that considers two evaluation measures, i.e., accuracy and G-mean, by optimizing a cost-sensitive Support Vector Machine and its hyper-parameter by a genetic algorithm. Experimental results on two benchmark datasets show that the proposed method is effective and efficient in comparison with the commonly used grid search method.