A novel imbalanced learning using dynamic multiclass evolutionary oversampling sigmoid deep neural network techniques

The majority and minority class data variably reduce the quality of the data produced and make the network harder to train for class-imbalanced datasets. It is uncertain how the minority class data are distributed and how they are input into the model. Furthermore, due to their significant bias in favour of the majority class, typical machine learning classifiers suffer considerably. To address these issues, we proposed multi‑class data stream learning using evolutionary and oversampling (EVOS) techniques. Evolutionary feature selection and effective oversampling combine for multi-class data stream learning. The aforementioned procedures enable the generator model to produce realistic-looking samples of both the majority and minority classes by learning the beneficial characteristics of each. According to the experimental findings of the proposed approach, classification performance significantly enhances when compared to alternative state-of-the-art methods.