Forensic Data Fusion Using Machine Learning for Disaster Victim Identification: Integrating DNA, Dental, and Image Data toward Resilient Urban Response (SDG 3 and SDG 11)

This research introduces a multimodal forensic data fusion framework powered by machine learning, aimed at improving Disaster Victim Identification (DVI) by integrating DNA, dental, and facial image information. Employing a synthetic dataset comprising 10,000 records (5,000 ante-mortem and 5,000 postmortem pairs), the study created and evaluated two fusion strategies, feature-level fusion and decision-level fusion to overcome the shortcomings of unimodal forensic techniques. Every data modality underwent preprocessing and standardization: DNA profiles included 30 Single Nucleotide Polymorphisms (SNPs), dental radiographs measured three primary odontometric parameters (tooth count, treatment, bone loss), and facial embeddings were condensed to 64-dimensional vectors utilizing a ResNet-50 CNN. Experimental assessment demonstrated significant performance enhancements for the fused models in contrast to single-modality baselines. Feature-level fusion reached an accuracy of 94.8%, with precision at 0.94, recall at 0.94, and a ROC-AUC of 0.97, whereas decision-level fusion achieved an accuracy of 93.5%. In comparison, unimodal classifiers achieved accuracies of 88.4% (DNA), 82.7% (dental), and 84.9% (image). The system exhibited strong resilience in the absence of certain modalities, showing an average accuracy decline of merely 2.5-4.3% and an average processing time of 0.60 seconds per record. Statistical analysis (ANOVA, F = 18.45, p = 0.0002) verified notable performance discrepancies between unimodal and fused models. The building collapse case study attained 94.8% accuracy with complete data and 88.7% with significantly degraded data, confirming its adaptability to actual disaster scenarios. Moreover, attention weight evaluations demonstrated fluctuating modality prioritization (DNA = 0.40, Dental = 0.34, Image = 0.26), emphasizing the importance of interpretability and decision transparency. The suggested framework advances SDG 3 (Good Health and Well-being) and SDG 11 (Sustainable Cities and Communities) by facilitating quicker, more precise, and ethically sound victim identification. These results set a new standard for data-driven forensic science, showing a 6-12% increase in accuracy and a 30% decrease in processing time relative to current systems.