Hybrid neural network with cost-sensitive support vector machine for class-imbalanced multimodal data.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Data Science

Background:

• Deep learning struggles with class-imbalanced multimodal data.
• Existing methods fail to adequately address skewed datasets.
• Multimodal data analysis requires robust solutions for imbalance.

Purpose of the Study:

• To propose a novel hybrid neural network with a cost-sensitive support vector machine (hybrid NN-CSSVM).
• To effectively address the class-imbalance problem in multimodal datasets.
• To improve classification performance on datasets with minority class underrepresentation.

Main Methods:

• Developed a fused multiple-network structure for multimodal feature extraction.
• Employed cost-sensitive support vector machines (SVMs) as the core classifier.
• Introduced a cost-sensitive SVM loss function with adjustable misclassification weights.

Main Results:

• The hybrid NN-CSSVM demonstrated excellent performance across various imbalance ratios.
• The model successfully classified data with minority class proportions as low as 2%.
• The approach effectively leverages complementary strengths of integrated architectures.

Conclusions:

• The hybrid NN-CSSVM is a powerful solution for class-imbalanced multimodal data.
• Cost-sensitive learning significantly enhances model robustness on skewed datasets.
• This method offers a promising direction for real-world multimodal classification challenges.