Iterative Framework For Data Augmentation Of Segmented Fingerprints

Researchers have proposed two novel data augmentation methods to address challenges in infant biometrics and chronic kidney disease (CKD) prediction.

One method uses iterative techniques to generate diverse variants of segmented infant fingerprints, addressing the scarcity of available data for research^[1]. Experiments on real infant fingerprints demonstrate the method's effectiveness in expanding fingerprint variability. Meanwhile, a separate study introduced BGCS, a two-stage data augmentation method tailored to binary clinical data for CKD prediction. BGCS generates synthetic minority-class samples using a Gaussian copula framework and applies a fine-tuned GPT-2 classifier to filter out clinically implausible samples. The EHR dataset used for CKD prediction contained 15,169 patients with CKD, collected between 2008 and 2022^[2]. BGCS outperformed other data augmentation methods, including SMOTE, CTGAN, and standard Gaussian Copula, in early dialysis prediction. The best-performing BGCS-augmented model was integrated into a decision tree-based clinical decision support system. The median minority-class recall of BGCS across classifiers was reported as 0.78^[2] and 0.87^[2], indicating varying performance across different classifiers.