DF-DM: A foundational process model for multimodal data fusion in the artificial intelligence era.

Area of Science:

• Data Mining and Machine Learning
• Multimodal Data Integration
• Computational Efficiency

Background:

• Integrating diverse data modalities presents significant challenges in big data, especially in healthcare.
• Existing Data Fusion models can be computationally expensive, complex, and prone to bias.
• Need for efficient and reliable methods for multimodal data analysis.

Approach:

• Introduced a novel process model for multimodal Data Fusion for Data Mining.
• Integrated embeddings and Cross-Industry Standard Process for Data Mining (CRISP-DM) with the Data Fusion Information Group (DFIG) model.
• Proposed 'disentangled dense fusion,' a new embedding fusion method to optimize mutual information and inter-modality feature interaction.

Key Points:

• The model aims to decrease computational costs, complexity, and bias while enhancing efficiency and reliability.
• Demonstrated efficacy in predicting diabetic retinopathy, domestic violence, and analyzing radiological data.
• Achieved high performance metrics: Macro F1 of 0.92, R-squared of 0.854, sMAPE of 24.868, and macro AUCs of 0.92 and 0.99.

Conclusions:

• The Data Fusion for Data Mining model significantly impacts multimodal data processing.
• The approach is suitable for diverse, resource-constrained settings.
• Highlights potential for broader adoption in various data-intensive fields.