Multimodal Data Fusion For Enhanced Predictive Modeling In AI Systems

Abstract

The growing demand for accurate and early prediction of chronic diseases requires advanced intelligent healthcare systems capable of handling heterogeneous medical data. This research proposes an artificial intelligence-based multimodal framework for chronic disease risk prediction through effective data fusion. The model integrates electrocardiogram (ECG) signals and clinical data, including hemodynamic parameters, laboratory biomarkers, and phenotypic attributes. ECG signals are preprocessed using bilateral filtering to remove noise while preserving structural details and clinical data are normalized using Min–Max normalization. Feature extraction is performed using Local Binary Pattern (LBP) for ECG signals and Principal Component Analysis (PCA) for clinical data. The extracted features are combined using a weighted and attention-based fusion strategy to effectively align multimodal representations. The fused features are processed using a Multimodal Convolutional Neural Network (MM-CNN), which learns hierarchical representations from integrated ECG and clinical data for effective cardiovascular risk prediction. The network captures complex intra- and inter-modality patterns through convolutional feature learning and attention-based fusion. To further enhance performance, a Quokka swarm optimization (QSO) strategy is applied to optimize model hyperparameters, improving convergence, robustness, and generalization. The entire model is implemented using Python. The proposed framework is evaluated on a multimodal cardiovascular dataset and demonstrates superior performance compared to conventional approaches, achieving an accuracy of 94.54%, precision of 92.67%, recall of 93.26%, and an F1-score of 94.64%, with an area under the curve exceeding 0.92. These results highlight the effectiveness of the proposed model in supporting early diagnosis and proactive clinical decision-making.