Big data has become easily available, but there is a need to improve the usefulness of these data, especially when we have an imbalanced dataset and overlapping data points in two or more classes. Machine-learning algorithms have improved in recent years, and many algorithms have been introduced that tackle the issues in data that su er from imbalanced classes and have overlap in some features. This will be a problem used to train a classi er in deciding where each data point belongs. Such a situation often occurs when the number of examples that we are interested in is much less in number than the other classes. We can see problems of this kind in many elds, like for example, fraud detection, cancer diagnosis, oil mining, network intrusion, and many others. In this thesis, we will discuss the cases of datasets that are imbalanced and overlapping in some data points. The main problem to be dealt with is how to make a better judgment regarding the gray area between the minority class and the majority class and the overlap between the two. We will provide characteristics of the imbalanced dataset scenarios in the classi cation phase and then try to provide a better solution. Then, we will discuss the cost of the learning process together with algorithms and techniques for solving these issues.

Doctor of Science (PhD)

Thesis