مقايسه روش‌هاي رده‌بندي داده‌هاي نامتوازن

One of the fundamental an‎d challenging issues in the field of machine learning is the problem of imbalanced data classification. In many data analysis appli-cations, such as fraudulent transaction detection, traffic accident analysis, rare disease identification, an‎d warning systems, the data is distributed in such a way that the number of samples in some classes is significantly smaller than the num-ber of samples in other classes. This imbalance causes conventional classification methods to perform poorly in identifying minority samples. The aim of this thesis is to conduct a comprehensive review an‎d comparison of various methods for clas: ig imbalanced data an‎d to propose a model for a more efficient solution to this problem. First, the fundamental concepts an‎d the importance of balanced data are explained, an‎d then the existing approaches to deal with this challenge are examined. These approaches include creating artificial balance between classes, modifying classification algorithms, an‎d using hybrid methods. The proposed model of this research, called DBC, is designed by combin- ing deep learning structures an‎d simultaneously applying both data-driven an‎d algorithm-driven approaches. With this model, it is possible to adjust the decision- making probability threshold, extract complex data features, an‎d identify minority class samples with higher accuracy. Subsequently, various classification models were implemented on three real- world imbalanced datasets, an‎d their performance was eva‎luated using diverse metrics. The obtained results show that by using hybrid approaches along with appropriate data preprocessing, the performance of models in solving the imbal- anced data classification problem can be improved.

5

147024