Classification of Imbalanced Data Represented as Binary Features

Typically, classification is conducted on a dataset that consists of numerical features and target classes. For instance, grayscale image, which usually represented as matrix integers varying from 0 to 255, enables one apply various algorithms image tasks. However, datasets binary cannot use many standard machine learning optimally, yet their amount not negligible. On the other hand, oversampling such synthetic minority technique (SMOTE) its variants are often used if for imbalanced. since SMOTE synthesize new samples based original samples, diversity synthesized highly limited due poor representation features. To solve this problem, preprocessing approach studied. By converting into ones using feature extraction methods, succeeding methods can fully display potential in improving classifiers’ performances. Through comprehensive experiments benchmark real medical datasets, it was observed converted consisting better (maximum improvements accuracy F1-score were 35.11% 42.17%, respectively). In addition, confirmed synergistically contribute improvement performance.