Synthetic Image Generation for Mitigating Overfitting in Deep Learning under Data-Scarce Conditions

This record contains the published version of the journal article “Synthetic Image Generation for Mitigating Overfitting in Deep Learning under Data-Scarce Conditions”, published in the International Journal of Innovative Research in Technology (IJIRT). The paper investigates lightweight, label-preserving synthetic image generation techniques to mitigate overfitting and improve generalization in deep learning models trained under extreme data-scarce conditions. Experiments conducted on the CIFAR-10 dataset using a ResNet-18 architecture demonstrate improved test accuracy, more stable training dynamics, and enhanced robustness without relying on complex generative models.