Hybrid Synthetic Minority Over-sampling Technique (HSMOTE) and Ensemble Deep Dynamic Classifier Model (EDDCM) for big...

Area of Science:

• Computer Science
• Machine Learning
• Data Science

Background:

• Big Data Classification (BDC) is crucial in healthcare, e-commerce, and banking.
• Conventional machine learning models struggle with high dimensionality and class imbalance in BDC.
• Existing methods often fail to effectively handle imbalanced datasets and select relevant features.

Purpose of the Study:

• To propose a hybrid framework enhancing Big Data Classification (BDC) effectiveness.
• To address challenges of high dimensionality and class imbalance in BDC.
• To improve the reliability and accuracy of classification models in complex datasets.

Main Methods:

• Introduced Hybrid Synthetic Minority Over-sampling Technique (HSMOTE) for class imbalance handling.
• Developed Optimization Ensemble Feature Selection Model (OEFSM) using meta-heuristic algorithms (FWDFA, AEHO, FWGWO) for robust feature selection.
• Proposed Ensemble Deep Dynamic Classifier Model (EDDCM) integrating DWCNN, DWBi-LSTM, and WAE with a dynamic ensemble strategy.

Main Results:

• The proposed framework demonstrated improved classification results across various datasets.
• Significant performance enhancements were observed particularly under conditions of class imbalance and high dimensionality.
• The integration of HSMOTE, OEFSM, and EDDCM effectively improved precision, recall, F-measure, and accuracy.

Conclusions:

• The hybrid framework offers a robust solution for Big Data Classification challenges.
• The combination of advanced sampling, feature selection, and deep learning ensemble methods enhances model performance.
• This approach provides a reliable method for improving classification accuracy in imbalanced and high-dimensional data.