Feature-to-Image Data Augmentation: Improving Model Feature Extraction with Cluster-Guided Synthetic Samples

TL;DR

This paper presents FICAug, a feature-to-image data augmentation method that enhances model generalization in low-resource settings by generating structured synthetic samples through clustering and neural network projection.

Contribution

FICAug introduces a novel framework combining feature clustering, Gaussian sampling, and neural network projection to improve data augmentation and model performance.

Findings

Achieved 84.09% cross-validation accuracy with feature augmentation.

Boosted ResNet-18 accuracy to 88.63% with reconstructed images.

Significantly improved classification performance in limited data scenarios.

Abstract

One of the growing trends in machine learning is the use of data generation techniques, since the performance of machine learning models is dependent on the quantity of the training dataset. However, in many real-world applications, particularly in medical and low-resource domains, collecting large datasets is challenging due to resource constraints, which leads to overfitting and poor generalization. This study introduces FICAug, a novel feature-to-image data augmentation framework designed to improve model generalization under limited data conditions by generating structured synthetic samples. FICAug first operates in the feature space, where original data are clustered using the k-means algorithm. Within pure-label clusters, synthetic data are generated through Gaussian sampling to increase diversity while maintaining label consistency. These synthetic features are then projected back into the image domain using a generative neural network, and a convolutional neural network is trained on the reconstructed images to learn enhanced representations. Experimental results demonstrate that FICAug significantly improves classification accuracy. In feature space, it achieved a cross-validation accuracy of 84.09%, while training a ResNet-18 model on the reconstructed images further boosted performance to 88.63%, illustrating the effectiveness of the proposed framework in extracting new and task-relevant features.