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

Abstract

In the big data era, integrating diverse data modalities poses significant challenges, particularly in complex fields like healthcare. This paper introduces a new process model for multimodal Data Fusion for Data Mining, integrating embeddings and the Cross-Industry Standard Process for Data Mining with the existing Data Fusion Information Group model. Our model aims to decrease computational costs, complexity, and bias while improving efficiency and reliability. We also propose "disentangled dense fusion," a novel embedding fusion method designed to optimize mutual information and facilitate dense inter-modality feature interaction, thereby minimizing redundant information. We demonstrate the model's efficacy through three use cases: predicting diabetic retinopathy using retinal images and patient metadata, domestic violence prediction employing satellite imagery, internet, and census data, and identifying clinical and demographic features from radiography images and clinical notes. The model achieved a Macro F1 score of 0.92 in diabetic retinopathy prediction, an R-squared of 0.854 and sMAPE of 24.868 in domestic violence prediction, and a macro AUC of 0.92 and 0.99 for disease prediction and sex classification, respectively, in radiological analysis. These results underscore the Data Fusion for Data Mining model's potential to significantly impact multimodal data processing, promoting its adoption in diverse, resource-constrained settings.

Keywords: Data Fusion; Embeddings; Foundational Models; Multimodal Data.

Fig. 1

DFGI Data Fusion model proposed, including AI and ML from [41]. The Levels where AI and ML techniques are proposed can be seen in red. The original DFGI model can be seen in blue.

Fig. 2

The proposed Data Fusion for Data Mining Model (DF-DM). The model is based on the DFGI model, integrating AI and ML but adding other functionalities vital for data mining tasks in orange.

Fig. 3

Illustrative Framework for Utilizing Foundational Models in Various Tasks. The figure assumes an initial foundational model for a general task and 3 different options. Option 1 is Zero-shot learning using the foundational model directly for a downstream task. Option 2 suggests the use of embedding, where embeddings of the original data are extracted and used for downstream task training. Option 3 means fine-tuning the full model for a specific task. The resulting model can also be used for embedding extraction.

Fig. 4

Disentangled dense data fusion model for classification tasks.

Fig. 5

Satellite image embedding extraction approach using a variational autoencoder with a Resnet 50 V2 backbone as encoder.

Fig. 6

Disentangled dense data fusion for temporal prediction tasks.