A synergistic fusion of shallow and deep generative model to enhance machine learning efficacy and classification performance in data-scarce environments

Data-constrained environments present a significant challenge to the effectiveness of machine learning and deep learning algorithms. The performance of these algorithms is inherently dependent on the quality and quantity of the training data they are exposed to. When training data is insufficient, the efficacy of downstream learning tasks is markedly diminished. The constrained availability and scarcity of data result from diverse intrinsic factors, encompassing data regulations, privacy concerns, the confidential nature of data, and the inherent rarity of data of interest in critical real-world applications.  To tackle this, researchers have come up with the concept of synthetic data to provide a low-cost, easily available and secure alternative. Synthetic data serves to bolster the robustness of model learning within real-world contexts, addressing the formidable challenge posed by data scarcity. This scarcity leads to class imbalance problem and insufficient representation of data very often encountered in real world datasets. Popular data generation strategies involve increasing the representation of minority class instances through the generation of synthetic examples. The existing data generation techniques aim to expand datasets for balancing, yet they frequently fall short in achieving satisfactory sample diversity. This paper examines the potential of shallow interpolation based data generation technique to capture the local dynamics for minority balancing and deep generative modeling based generative adversarial networks (GANs) for global distribution estimation for augmentation the smaller datasets. This paper presents a hybrid amalgamation approach for tabular data involving mixed type attributes and pays special attention to data imbalance and insufficient data problems. We named this approach as hybrid data balancing and augmentation approach on mixed tabular data (HDBA-MTD) tailored for synthesizing samples on underrepresented target labels (skewed class labels) and insufficient data instances. This approach exhibits the potential to restore dataset balance, address bias in the dataset, mitigate over-fitting issues, enhance training data diversity, thereby pays special attention towards the downstream classification and generalization performance in data rarity dilemmas. Experiments are conducted on benchmark datasets to validate the feasibility of the proposed model in realistic scenarios. The evaluation and analysis of experimental procedures demonstrate favorable comparisons with other existing synthetic data generation techniques.