Imbalanced Multiclass Data Classification Using Ant Colony Optimization Algorithm

Class imbalance problems have drawn increasing interest lately because of its classification trouble caused by imbalanced class deliveries and poor prediction performance for minority class. This problem is particularly common in preparation and can be detected in various disciplines including fraud detection, anomaly detection, oil spillage detection, medical diagnosis, facial recognition. Many ensemble procedures only concentrated on two-class imbalance problems. There are numerous unresolved concerns in multiclass imbalanced problems. Using One-vsOne binarization technique for disintegrating the original multiclass data-set into binary classification problems. Then, each and every time these binary sub problems is imbalanced, applying undersampling step, using the ACOsampling algorithm in order to rebalance the data. Only taking out high frequency dataset from majority samples and mingling those with all minority samples to build the final balanced training set. Here taken different multiclass data such as thyroid, lung cancer and contraceptive. Finally evaluate the performance of each method on four benchmark skewed DNA microarray dataset by support vector machine (SVM) Classifier. It gives better accuracy, precision, f-measure and g-mean when comparing with RUS (Random Under Sampling) method.