Evaluating Fidelity and Machine Learning Utility of Synthetic Tabular Data Generated Using Generative Models

Evaluating Fidelity and Machine Learning Utility of Synthetic Tabular Data Generated Using Generative Models | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Evaluating Fidelity and Machine Learning Utility of Synthetic Tabular Data Generated Using Generative Models Aaditya Kumar Dhaka, Apash Roy, S Shrivallabha This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7287372/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 8 You are reading this latest preprint version Abstract Synthetic tabular data offers a promising solution for enabling privacy-preserving machine learning in sensitive domains such as healthcare. However, assessing the fidelity and utility of such data remains challenging. In this study, we evaluate four generative models—CTGAN, TVAE, Gaussian Copula, and CopulaGAN—on a benchmark dataset for stroke prediction. We propose a two-phase generation and evaluation framework that combines statistical diagnostics with feature-level fidelity analysis and downstream classification performance. Our findings highlight significant variation across models, with TVAE and Gaussian Copula achieving superior fidelity and generalization. The results demonstrate that high structural similarity does not always guarantee practical machine learning utility. Synthetic Data Tabular Data Generation CT- GAN TVAE Gaussian Copula Machine Learning Evaluation Data Fidelity Figures Figure 1 Figure 2 Figure 3 Figure 4 I. INTRODUCTION Machine learning (ML) is increasingly applied in sensitive domains such as healthcare, finance, and human resources. However, the availability of high-quality, diverse, and privacy- compliant datasets remains a persistent challenge. Regulatory frameworks such as the General Data Protection Regulation (GDPR) and the Health Insurance Portability and Accountabil- ity Act (HIPAA) restrict access to real-world data, particularly in healthcare settings [ 1 ]. This constraint impacts not only research reproducibility but also the development of general- izable ML models. As a result, synthetic data generation has gained prominence as a privacy-preserving alternative for data sharing and downstream ML experimentation [ 2 , 13 ]. Among synthetic data types, synthetic tabular data holds particular importance due to its relevance in structured elec- tronic health records and risk prediction systems [ 4 , 5 ]. Syn- thetic Tabular Data Generation (STDG) aims to reproduce the statistical properties of real data while eliminating identifiable patient-level information. A wide range of models—ranging from statistical copulas to deep generative approaches like GANs and VAEs—have been proposed for this task [ 12 , 18 ]. These models differ in stability, interpretability, and perfor- mance, and often lack standard evaluation frameworks. Despite recent advances, evaluating the quality and practical utility of synthetic tabular data remains an open research problem. While tools such as SDMetrics provide structural and distributional diagnostics [ 7 ], they often fail to detect deeper issues such as class-level divergence, fairness shifts, or mode collapse [ 20 , 9 ]. Moreover, most metrics report only aggregate scores, limiting interpretability at the column level and offering limited guidance for model selection in real-world contexts. This study addresses these challenges through a practical and interpretable framework for evaluating synthetic tabular data. We benchmark four widely used models—CTGAN, TVAE, Gaussian Copula, and CopulaGAN—on a publicly available benchmark stroke dataset. Our key contributions include: A two-phase training and generation strategy for stable and optimized synthesis. A column-wise fidelity framework using class-conditional deviation and Total Variation Distance (TVD) to assess both numerical and categorical features. A comprehensive evaluation of synthetic data using SD- Metrics and downstream classification via ML models such as Random Forest and XGBoost on real hold-out data. An analysis of how statistical fidelity correlates with ML utility, highlighting model strengths and trade-offs. The rest of the paper is organized as follows: Section II reviews related work in synthetic data generation and evaluation. Section III outlines our dataset, generative models, and evaluation methodology. Section IV presents the results. Section V discusses key findings. Section VI outlines limita- tions and future directions. Section VII concludes the paper. II. RELATED WORK Synthetic Tabular Data Generation (STDG) has emerged as a critical research area due to increasing privacy constraints and the demand for shareable, machine learning-compatible datasets in sensitive domains such as healthcare. This section reviews existing work grouped into major technical themes. A. Statistical and Copula-Based Models Early STDG methods focused on probabilistic modeling, such as Gaussian Copulas and Bayesian networks. These models capture marginal and pairwise dependencies effec- tively, and are computationally efficient. Figueira and Vaz [ 2 ] categorized such techniques under rule-based and statistical methods. Wang et al. [ 19 ] introduced hybrid frameworks combining copula structures with neural networks. Herurkar et al. [ 20 ] proposed novel fidelity metrics (FAED, FPCAD) that address shortcomings of aggregate metrics in capturing structural and class-conditional variation. B. GAN and VAE-Based Generative Models GANs and VAEs remain dominant in deep generative mod- eling. Hernandez et al. [ 1 ] reviewed GAN-based models in healthcare and noted their instability and poor mode coverage. Manousakas and Aydo¨re [ 4 ] benchmarked CTGAN, TVAE, and Gaussian Copula, highlighting that ML utility depends heavily on data complexity. Panagiotou et al. [ 5 ] and Stoian et al. [ 9 ] found that GAN-based generators often suffer from class imbalance and fairness distortions. CopulaGAN and CTGAN have shown promise in specific tabular tasks [ 14 , 16 ], though performance varies significantly by dataset. C. Transformer and Diffusion-Based Methods Recent innovations explore attention-based and diffusion models for tabular generation. Ma et al. [ 13 ] introduced TabPFGen, a transformer-based model that performs well on small datasets. Kim et al. [ 17 ] and Wang et al. [ 19 ] surveyed these newer architectures, noting their expressive power but also higher training instability and reduced interpretability. D. Evaluation Metrics and ML Utility Assessment SDMetrics is the most commonly used diagnostic suite [ 7 ], offering shape, trend, and structure metrics. However, it lacks interpretability and fails to capture class-level fidelity breakdowns. Rashidi et al. [ 6 ] combined SDMetrics with AutoML pipelines. Herurkar et al. [ 20 ] argue that metrics like FAED better capture deviations. Other studies [ 12 , 15 ] use Kolmogorov-Smirnov, TVD, and Spearman correlation to assess fidelity. ML utility is often tested using classifiers (SVM, LR, RF) and metrics like Accuracy, F1-score, and ROC AUC [ 16 , 18 ]. However, these are sensitive to data imbalance and noise. E. Research Gaps Despite growing progress, prior work still lacks: Interpretable, feature-level diagnostics of fidelity; Strong correlation analysis between fidelity and ML util- ity. Our study addresses these gaps by benchmarking four pop- ular STDG models—CTGAN, TVAE, Gaussian Copula, and CopulaGAN—on a benchmark stroke prediction dataset. We introduce a two-phase generation and evaluation framework that combines SDMetrics, column-wise fidelity analysis using TVD, and downstream machine learning performance evalua- tion. Although both Random Forest and XGBoost classifiers were initially tested, Random Forest was retained for final analysis due to comparable performance. III. METHODOLOGY This section outlines the dataset, generative models, training setup, and evaluation framework used in this study. Our approach combines statistical and practical assessment metrics to evaluate synthetic tabular data fidelity and utility. A. Dataset Description We use a publicly available benchmark dataset for stroke prediction, originally based on the Kaggle dataset by Sori- ano [ 29 ]. For modeling purposes, we utilized a synthetically expanded version of this dataset containing 15,000 samples and 22 features. The dataset includes a mix of numerical and categorical variables relevant to stroke risk, along with a binary target variable indicating stroke occurrence. After preprocessing, the final dataset used in this study consists of 8 numerical and 13 categorical features (including the target variable). The target classes are balanced in the dataset. B. Generative Models Evaluated We benchmark four popular generative models: CTGAN – A GAN-based tabular synthesizer that uses conditional vectors to handle categorical data. TVAE – A VAE-based model designed for capturing nonlinear latent representations in tabular data. Gaussian Copula (GC) – A statistical model using fitted marginals and a Gaussian copula to capture inter-feature dependencies. CopulaGAN – A hybrid model combining copula trans- formations with adversarial training. C. Two-Phase Training Strategy We use a two-phase generation framework: Phase 1: Epoch Tuning – Each deep model is trained on a 2,000-row subset. Multiple epoch settings are evaluated (200 to 500), and optimal settings are selected based on fidelity scores. Phase 2: Full Generation – Final models are trained on the full 15,000-row dataset using the selected epochs. Each model generates three synthetic datasets of sizes 5K, 15K, and 25K. For GC and CopulaGAN, marginal distributions for numer- ical features are fitted using the fitter package. Best-fit distributions are selected based on minimizing the sum of squared errors. D. Evaluation Framework 1. Column-Wise Fidelity Analysis : We propose a feature- level diagnostic approach based on class-conditional deviation. For a numerical feature X with classes C 1 and C 2 , we compute the absolute difference between real and synthetic means per class: 3. Machine Learning Utility Evaluation: To assess practi- cal utility, we trained both a Random Forest and an XGBoost classifier on synthetic data and evaluated them on a 5,000-row real hold-out set. Since both models produced highly similar results across all metrics, we present Random Forest perfor- mance for clarity and consistency. Performance is measured using: Sensitivity (Recall for positive class) Specificity (Recall for negative class) Balanced Accuracy ROC AUC This simulates real-world use cases where synthetic data is used to train ML models later applied to real environments. IV. RESULTS This section presents fidelity and machine learning utility results for synthetic datasets generated using four generative models: CTGAN, TVAE, Gaussian Copula, and CopulaGAN. Evaluation includes SDMetrics scores, column-wise fidelity analysis, class distribution matching, and downstream classi- fication performance on real data. A. Epoch Tuning and Model Selection Epoch tuning was performed for the three deep generative models (CTGAN, TVAE, and CopulaGAN) using a 2,000- row subset. SDMetrics scores and our custom column-wise diagnostic framework were used to select the optimal epoch. Table I summarizes the results. TABLE I: Epoch Tuning Summary for Deep Generative Models B. Synthetic Data Generation and Selection Each model was trained using its optimal epoch on the full 15,000-row real dataset. To ensure a fair comparison, we gen- erated 10 synthetic datasets of the same size (15,000 rows) for each model. The final dataset was selected using our column-wise diagnostic framework, prioritizing the highest number of Super and Moderate columns and the lowest number of Worst columns. In the case of ties, the dataset with the lowest average feature-level deviation was chosen. Table II summarizes the column fidelity and average devi- ation for the selected dataset from each model. TABLE II: Final Selected Synthetic Dataset Per Model (15K rows) As shown in Table II, the Gaussian Copula model achieved perfect fidelity labeling with all 21 features rated as Super and the lowest average feature deviation (0.0193), indicating strong statistical alignment with the real data. In contrast, TVAE exhibited higher deviation and more Worst-rated features, although it later demonstrated strong downstream classification performance. To further illustrate these results, Figure 2 shows the break- down of Super, Moderate, and Worst feature counts for each selected synthetic dataset. This visual comparison reinforces the sharp contrasts in column-wise fidelity across models. C. SDMetrics Fidelity Evaluation We use the SDMetrics library from the SDV framework to evaluate statistical fidelity of each synthetic dataset (5k, 15k, 25k) acorss all four models. Two official reports were computed: Diagnostic Report : Assesses schema compliance, null value constraints, and structural issues in the synthetic dataset. Quality Report : Measures column-wise distributional similarity and pairwise feature relationships, summarized as a Data Quality Score. The reports were computed using the run diagnostic() and evaluate quality() functions respectively from the SDMetrics library. Table III summarizes the resulting scores only for 15k dataset genereted by each of the four deep generative models. TABLE III: SDMetrics Scores for Final Synthetic Datasets (15k rows) Model Diagnostic Score (%) Data Quality Score (%) CTGAN 100 85.44 TVAE 100 71.64 Gaussian Copula 100 98.64 CopulaGAN 100 86.50 All models scored 100% in diagnostic validity, while the Gaussian Copula model achieved the highest overall data qual- ity. TVAE, despite a slightly lower quality score, performed better in downstream ML utility as we will see later. D. Target Class Distribution Comparison To assess whether generative models preserved the original outcome distribution, we analyzed the class proportions of the target variable Diagnosis (stroke vs. no-stroke) in each 15K-row synthetic dataset. The original dataset exhibited near- perfect balance between the two classes (approximately 50.2% no-stroke and 49.8% stroke). Figure 3 illustrates the distribution of class labels across the real dataset and synthetic datasets generated by CTGAN, TVAE, Gaussian Copula, and CopulaGAN. While all models preserved the general structure of the binary class distribution, Gaussian Copula achieved the closest match to the original dataset (within 0.3%). TVAE over- represented the stroke class (by approximately 9%), while CT- GAN and CopulaGAN showed moderate under-representation of stroke cases. This alignment with class distribution is important for minimizing bias in downstream classification performance. E. Machine Learning Utility Evaluation : To assess whether synthetic data supports downstream predictive tasks, we trained a Random Forest classifier on each synthetic dataset and evaluated it on a 5,000- row real hold-out set. The classifier was trained us- ing 10-fold stratified cross-validation with default hy- perparameters (n_estimators=100, max_depth=None, random_state=42). Table IV reports the mean values across four evaluation metrics: Sensitivity, Specificity, Balanced Accuracy, and ROC AUC. For fair comparison, we include results for the real dataset at three sizes (5K, 15K, 25K) alongside synthetic datasets of equal sizes generated by each model. TABLE IV: ML Utility Metrics (Mean over 10-Fold CV): RF Trained on Synthetic, Tested on Real Model Size Sensitivity Specificity Bal. Accuracy ROC AUC Original (Real) 5K 0.4402 0.5165 0.4784 0.4829 15K 0.4514 0.5356 0.4935 0.4925 25K 0.4587 0.5273 0.4930 0.4993 TVAE 5K 0.8652 0.6768 0.7710 0.8613 15K 0.8781 0.7158 0.7970 0.8939 25K 0.8853 0.7283 0.8068 0.9012 CTGAN 5K 0.0548 0.9714 0.5131 0.5819 15K 0.0415 0.9831 0.5123 0.5864 25K 0.0640 0.9704 0.5172 0.5887 Gaussian Copula 5K 0.5642 0.4407 0.5024 0.5073 15K 0.4262 0.5730 0.4996 0.5016 25K 0.4756 0.5282 0.5019 0.5011 CopulaGAN 5K 0.0482 0.9769 0.5126 0.5798 15K 0.0539 0.9696 0.5118 0.5772 25K 0.0452 0.9768 0.5110 0.5853 Across all models, TVAE demonstrated the strongest generalization ability, consistently outperforming other synthesizers and even the original dataset. Its performance improved with sample size, achieving the highest ROC AUC (0.9012) and balanced accuracy (0.8068) on the 25K dataset. This suggests strong preservation of class boundaries in the synthetic data. CTGAN and CopulaGAN, while yielding high specificity, suffered from extremely low sensitivity. These results suggest overfitting to the majority class and confirm observations from earlier diagnostics—particularly the imbalanced target class distribution. Gaussian Copula achieved moderate performance, with relatively balanced sensitivity and specificity, aligning with its strong feature-wise fidelity but limited downstream utility. Figure 4 provides a visual summary of the classification performance for 15K datasets across all models compared to the real data baseline. V. DISCUSSION Our evaluation reveals notable differences in how each synthetic data generation model balances fidelity and machine learning utility. TVAE consistently delivered the strongest classification performance across all dataset sizes, achieving the highest ROC AUC and balanced accuracy. This suggests that, despite moderate statistical fidelity in terms of feature-level deviations, TVAE effectively captures class-separating patterns in the data. Its robustness to increasing dataset size further reinforces its utility for downstream predictive tasks. Gaussian Copula, on the other hand, demonstrated superior statistical fidelity—achieving the highest number of Super columns and the lowest average feature-wise deviation. How- ever, its classification performance remained modest. This trade-off suggests that high fidelity in marginal and condi- tional distributions does not necessarily translate to improved learning of decision boundaries, particularly in nonlinear tasks. CTGAN and CopulaGAN, while structurally sound accord- ing to SDMetrics, exhibited poor classification utility. Both models suffered from extremely low sensitivity and skewed target class distributions, likely due to mode collapse and over- fitting to the majority class during adversarial training. Their failure to generalize downstream highlights the limitations of using aggregate fidelity scores alone to judge synthesis quality. To make this contrast more explicit, Table V summarizes the relative ranking of each model based on fidelity and utility. TABLE V: Model Ranking: Statistical Fidelity vs. ML Utility Model Fidelity Rank Utility Rank TVAE 4 1 CTGAN 3 4 Gaussian Copula 1 2 CopulaGAN 2 3 As seen in Table V, the highest fidelity score was achieved by Gaussian Copula, while TVAE dominated utility-based evaluation. The two deep generative models, CTGAN and CopulaGAN, lagged in both aspects. This divergence further reinforces the need to assess synthetic data not just through global similarity metrics, but through task-oriented validation aligned with intended applications. Overall, our findings emphasize that fidelity and utility are not always aligned. Models like TVAE that may rank lower in statistical diagnostics can outperform others in real-world ML tasks. These results underline the importance of application- specific evaluation pipelines that combine both diagnostic and task-based assessments when benchmarking synthetic data generators. These findings motivate the need for more interpretable, utility-aware evaluation frameworks for synthetic data, which we further discuss in the next section. VI. LIMITATIONS AND FUTURE WORK While our evaluation framework provides detailed insights into fidelity and machine learning utility across multiple synthetic data generators, several limitations should be noted. First, the study focuses on a single healthcare dataset. While the results are indicative, generalizability to other domains (e.g., finance, marketing, social sciences) may be limited. Model behaviors—particularly those involving marginal fi- delity and class preservation—can vary across domains with different statistical properties or class imbalances. Second, while our main downstream utility evaluation is reported using a Random Forest classifier due to its strong performance and interpretability, we also experimented with XGBoost and observed similar results. However, other models such as neural networks might yield different utility scores. Incorporating a broader set of classifiers or model-agnostic evaluation methods could provide a more comprehensive as- sessment of synthetic data utility. Third, our fidelity evaluation focused on feature-wise class- conditional statistics and SDMetrics scores, which offer valu- able interpretability and standardized benchmarks. However, these metrics may not fully capture complex aspects such as higher-order correlations, fairness, privacy leakage, or tempo- ral consistency—dimensions that are increasingly important in real-world, high-stakes applications. Moreover, the selection of synthetic datasets was guided by column-level deviation thresholds to maintain transparency and tractability. While effective for identifying localized de- viations, this method may not account for global structural patterns or multi-feature interactions that could influence downstream model behavior. Finally, the evaluation was done using 10-fold cross- validation on a fixed real hold-out set. Though this setup is reproducible, future work can explore multiple test sets, adaptive validation splits, or synthetic-to-synthetic evaluation loops to assess model robustness further. In future work, we aim to extend this framework to diverse datasets and classifiers, incorporate fairness and privacy met- rics, and develop a unified scoring system that aligns statistical fidelity with downstream utility in a domain-aware manner. VII. CONCLUSION This study proposed a comprehensive and interpretable evaluation framework for assessing both statistical fidelity and machine learning utility of synthetic tabular data. Us- ing a benchmark dataset for stroke prediction, we evaluated four popular generative models—TVAE, CTGAN, Gaussian Copula, and CopulaGAN—across diagnostic, statistical, and downstream performance metrics. Our results demonstrate that fidelity alone is not a re- liable indicator of utility. While Gaussian Copula achieved the highest statistical similarity to the original data, TVAE outperformed all other models in ML utility across sensitivity, specificity, balanced accuracy, and ROC AUC. Conversely, CT- GAN and CopulaGAN struggled to preserve target distribution and showed poor generalization. These findings highlight the need for application-specific evaluation pipelines that move beyond global scores and incor- porate column-level diagnostics and predictive performance. Our framework offers a practical approach for researchers and practitioners to select suitable synthetic data generators based on intended downstream tasks. Declarations Author Contribution A.K.D. and A.R. jointly conceived the research idea and designed the overall methodology of the study. A.K.D. led the development of the experimental pipeline, including model implementation, training, testing, and synthetic data generation. S.S. contributed to the evaluation and interpretation of results, particularly focusing on statistical diagnostics and machine learning utility analysis. A.K.D. also prepared the manuscript draft. All authors contributed to editing and revising the manuscript and approved the final version for submission. Acknowledgement The authors would like to thank CHRIST (Deemed to be University), Bangalore, for providing institutional support throughout the research. We also extend our sincere gratitude to the Dr. Saravanan V (Scientist Grade -D), Indian Council of Medical Research (ICMR) for enabling this work as part of a student research internship. References M. Hernandez, M. T. Perez, L. Serra, and L. J. Garcia, “Synthetic data generation for tabular health records: A systematic review,” Neurocom- puting , vol. 493, pp. 28–45, 2022. A. Figueira and B. Vaz, “Survey on synthetic data generation, evaluation methods and GANs,” Mathematics , vol. 10, no. 15, p. 2733, 2022. E. Papadaki, A. G. Vrahatis, and S. Kotsiantis, “Exploring innovative approaches to synthetic tabular data generation,” Electronics , vol. 13, no. 10, p. 1965, 2024. D. Manousakas and S. Aydo¨re, “On the usefulness of synthetic tabular data generation,” arXiv preprint arXiv:2306.15636 , 2023. E. Panagiotou, A. Roy, and E. Ntoutsi, “Synthetic tabular data generation for class imbalance and fairness: A comparative study,” arXiv preprint arXiv:2409.05215 , 2024. J. Ma, H. Zhang, A. R. Mohamed, and M. Welling, “TabPFGen: Tabular data generation with TabPFN,” arXiv preprint arXiv:2406.05216 , 2024. H. H. Rashidi et al., “A novel and fully automated platform for synthetic tabular data generation and validation,” Scientific Reports , vol. 14, no. 1, p. 23312, 2024. R. Shi et al., “A comprehensive survey of synthetic tabular data generation,” arXiv preprint arXiv:2504.16506 , 2025. M. C. Stoian et al., “How realistic is your synthetic data? Con- straining deep generative models for tabular data,” arXiv preprint arXiv:2402.04823 , 2024. D. Panfilo et al., “A deep learning-based pipeline for the generation of synthetic tabular data,” IEEE Access , vol. 11, pp. 63306–63323, 2023. A. X. Wang et al., “Challenges and opportunities of generative models on tabular data,” Applied Soft Computing , vol. 141, p. 112223, 2024. D.-K. Kim, Y. Ryu, H. Lee, and H. Choi, “Generative models for tabular data: A review,” Journal of Mechanical Science and Technology , vol. 38, no. 9, pp. 4989–5005, 2024. C. Hassan, R. Salomone, and K. Mengersen, “Deep generative mod- els, synthetic tabular data, and differential privacy: An overview and synthesis,” arXiv preprint arXiv:2307.15424 , 2023. T. Liu et al., “Goggle: Generative modelling for tabular data by learning relational structure,” in Proc. Int. Conf. Learn. Representations (ICLR) , 2023. L. Xu and K. Veeramachaneni, “Synthesizing tabular data using gener- ative adversarial networks,” arXiv preprint arXiv:1811.11264 , 2018. K. Liu and R. B. Altman, “Conditional generative models for synthetic tabular data: Applications for precision medicine and diverse represen- tations,” Annu. Rev. Biomed. Data Sci. , vol. 8, 2025. B. Wen et al., “Causal-TGAN: Generating tabular data using causal gen- erative adversarial networks,” arXiv preprint arXiv:2104.10680 , 2021. A. X. Wang, B. P. Nguyen, and C. R. Simpson, “Generative AI for tabular data synthesis,” in Proc. Pacific-Asia Conf. Knowl. Discov. Data Mining (PAKDD) , Springer, 2025. X. Wen et al., “From supervised to generative: A novel paradigm for tabular deep learning with large language models,” in Proc. ACM SIGKDD Conf. Knowl. Discov. Data Mining , 2024. D. Herurkar, A. Ali, and A. Dengel, “Evaluating generative models for tabular data: Novel metrics and benchmarking,” arXiv preprint arXiv:2504.20900 , 2025. A. Gonzales, G. Guruswamy, and S. R. Smith, “Synthetic data in health care: A narrative review,” PLOS Digital Health , vol. 2, no. 1, e0000082, 2023. P. Davis, R. Lay-Yee, and J. Pearson, “Using micro-simulation to create a synthesised data set and test policy options: The case of health service effects under demographic ageing,” Health Policy , vol. 97, no. 2–3, pp. 267–274, 2010. C. Ngufor, H. Van, B. S. Caffo, N. D. Shah, and R. G. McCoy, “Mixed effect machine learning: A framework for predicting longitudinal change in hemoglobin A1c,” Journal of Biomedical Informatics , vol. 89, pp. 56–67, 2019. W. T. Enanoria et al., “The effect of contact investigations and public health interventions in the control and prevention of measles transmis- sion: A simulation study,” PLOS ONE , vol. 11, no. 12, e0167160, 2016. T. Laderas et al., “Teaching data science fundamentals through realistic synthetic clinical cardiovascular data,” bioRxiv , preprint 232611, 2017. SMART Health IT, “SMART on FHIR developer sandbox documenta- tion,” 2017. Centers for Disease Control and Prevention (CDC), “Public-use linked mortality file: 2015 linked mortality file description,” Mar. 2020. SDV Documentation, “Single Table Synthesizers,” SDV Dev Documen- tation. [Online]. Available: https://docs.sdv.dev/sdv F. Soriano, “Stroke Prediction Dataset,” Kaggle, 2021. [Online]. Available: https://www.kaggle.com/datasets/fedesoriano/ stroke-prediction-dataset Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Reviews received at journal 05 May, 2026 Reviewers agreed at journal 03 May, 2026 Reviewers agreed at journal 03 May, 2026 Reviewers agreed at journal 10 Sep, 2025 Reviewers invited by journal 10 Sep, 2025 Editor assigned by journal 10 Sep, 2025 Submission checks completed at journal 05 Aug, 2025 First submitted to journal 04 Aug, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7287372","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":515083768,"identity":"c3fdd8a2-19f2-435a-910c-4e9dda97e37d","order_by":0,"name":"Aaditya Kumar Dhaka","email":"","orcid":"","institution":"CHRIST (Deemed to be University)","correspondingAuthor":false,"prefix":"","firstName":"Aaditya","middleName":"Kumar","lastName":"Dhaka","suffix":""},{"id":515083769,"identity":"17c39cbe-5ef1-4262-87a0-0dfc2dea39e7","order_by":1,"name":"Apash Roy","email":"data:image/png;base64,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","orcid":"","institution":"CHRIST (Deemed to be University)","correspondingAuthor":true,"prefix":"","firstName":"Apash","middleName":"","lastName":"Roy","suffix":""},{"id":515083776,"identity":"62cdcad2-b07c-46c3-a7d7-1cce55acfa56","order_by":2,"name":"S Shrivallabha","email":"","orcid":"","institution":"CHRIST (Deemed to be University)","correspondingAuthor":false,"prefix":"","firstName":"S","middleName":"","lastName":"Shrivallabha","suffix":""}],"badges":[],"createdAt":"2025-08-04 06:09:13","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7287372/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7287372/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":91540343,"identity":"6e62b6f6-d8e3-434e-86bd-8c8f873ced38","added_by":"auto","created_at":"2025-09-17 13:49:20","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":216008,"visible":true,"origin":"","legend":"\u003cp\u003eTwo-phase training and evaluation pipeline for syn- thetic tabular data generation.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-7287372/v1/c1666c54f2f4c8ccc59dbdfa.png"},{"id":91539857,"identity":"073bc4c3-a0f7-42d7-9e08-8fe58e8819de","added_by":"auto","created_at":"2025-09-17 13:41:20","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":57139,"visible":true,"origin":"","legend":"\u003cp\u003eFeature fidelity labels (Super, Moderate, Worst) across selected synthetic datasets (15K rows).\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-7287372/v1/0c1901312dab0ab8244b12b2.png"},{"id":91541624,"identity":"ba9c1ab3-ecf2-47e1-a62b-6e558eb22118","added_by":"auto","created_at":"2025-09-17 13:57:20","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":30660,"visible":true,"origin":"","legend":"\u003cp\u003eTarget class distribution (\u003cem\u003eDiagnosis\u003c/em\u003e) in the real dataset and 15K synthetic datasets.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-7287372/v1/6a9cbb7fb3b13c7c470ffdb9.png"},{"id":91539860,"identity":"ea8b55ac-4c54-42f5-a2f3-da298af89d8c","added_by":"auto","created_at":"2025-09-17 13:41:20","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":33766,"visible":true,"origin":"","legend":"\u003cp\u003eML utility metrics across synthetic and real datasets (15K rows). TVAE outperformed others across all four metrics, followed by Gaussian Copula.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-7287372/v1/bd91bd25666c5deec3b8a88d.png"},{"id":91542902,"identity":"6b77c8ea-928a-4d15-8f63-779a48c4580b","added_by":"auto","created_at":"2025-09-17 14:13:20","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1014247,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7287372/v1/9385a62c-4cf6-4ada-a735-c9e872a6d0b8.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Evaluating Fidelity and Machine Learning Utility of Synthetic Tabular Data Generated Using Generative Models","fulltext":[{"header":"I.\tINTRODUCTION","content":"\u003cp\u003eMachine learning (ML) is increasingly applied in sensitive domains such as healthcare, finance, and human resources. However, the availability of high-quality, diverse, and privacy- compliant datasets remains a persistent challenge. Regulatory frameworks such as the General Data Protection Regulation (GDPR) and the Health Insurance Portability and Accountabil- ity Act (HIPAA) restrict access to real-world data, particularly in healthcare settings [\u003cspan class=\"CitationRef\"\u003e1\u003c/span\u003e]. This constraint impacts not only research reproducibility but also the development of general- izable ML models. As a result, synthetic data generation has gained prominence as a privacy-preserving alternative for data sharing and downstream ML experimentation [\u003cspan class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e13\u003c/span\u003e].\u003c/p\u003e\n\u003cp\u003eAmong synthetic data types, \u003cem\u003esynthetic tabular data\u003c/em\u003e holds particular importance due to its relevance in structured elec- tronic health records and risk prediction systems [\u003cspan class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e5\u003c/span\u003e]. Syn- thetic Tabular Data Generation (STDG) aims to reproduce the statistical properties of real data while eliminating identifiable patient-level information. A wide range of models\u0026mdash;ranging from statistical copulas to deep generative approaches like GANs and VAEs\u0026mdash;have been proposed for this task [\u003cspan class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e18\u003c/span\u003e]. These models differ in stability, interpretability, and perfor- mance, and often lack standard evaluation frameworks.\u003c/p\u003e\n\u003cp\u003eDespite recent advances, evaluating the quality and practical utility of synthetic tabular data remains an open research problem. While tools such as SDMetrics provide structural and distributional diagnostics [\u003cspan class=\"CitationRef\"\u003e7\u003c/span\u003e], they often fail to detect deeper\u003c/p\u003e\n\u003cp\u003eissues such as class-level divergence, fairness shifts, or mode collapse [\u003cspan class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e9\u003c/span\u003e]. Moreover, most metrics report only aggregate scores, limiting interpretability at the column level and offering limited guidance for model selection in real-world contexts.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eThis study addresses these challenges\u003c/strong\u003e through a practical and interpretable framework for evaluating synthetic tabular data. We benchmark four widely used models\u0026mdash;CTGAN, TVAE, Gaussian Copula, and CopulaGAN\u0026mdash;on a publicly available benchmark stroke dataset. Our key contributions include:\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003e\n \u003cp\u003eA two-phase training and generation strategy for stable and optimized synthesis.\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003eA column-wise fidelity framework using class-conditional deviation and Total Variation Distance (TVD) to assess both numerical and categorical features.\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003eA comprehensive evaluation of synthetic data using SD- Metrics and downstream classification via ML models such as Random Forest and XGBoost on real hold-out data.\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003eAn analysis of how statistical fidelity correlates with ML utility, highlighting model strengths and trade-offs.\u003c/p\u003e\n \u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eThe rest of the paper is organized as follows: Section\u003c/p\u003e\n\u003cp\u003eII reviews related work in synthetic data generation and evaluation. Section III outlines our dataset, generative models, and evaluation methodology. Section IV presents the results. Section V discusses key findings. Section VI outlines limita- tions and future directions. Section VII concludes the paper.\u003c/p\u003e"},{"header":"II.\tRELATED WORK","content":"\u003cp\u003eSynthetic Tabular Data Generation (STDG) has emerged as a critical research area due to increasing privacy constraints and the demand for shareable, machine learning-compatible datasets in sensitive domains such as healthcare. This section reviews existing work grouped into major technical themes.\u003c/p\u003e\n\u003cp\u003e\u003cspan\u003e\u003c/span\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eA. Statistical and Copula-Based Models\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003c/p\u003e\n\u003cp\u003eEarly STDG methods focused on probabilistic modeling, such as Gaussian Copulas and Bayesian networks. These models capture marginal and pairwise dependencies effec- tively, and are computationally efficient. Figueira and Vaz [\u003cspan class=\"CitationRef\"\u003e2\u003c/span\u003e] categorized such techniques under rule-based and statistical methods. Wang et al. [\u003cspan class=\"CitationRef\"\u003e19\u003c/span\u003e] introduced hybrid frameworks\u003c/p\u003e\n\u003cp\u003ecombining copula structures with neural networks. Herurkar et al. [\u003cspan class=\"CitationRef\"\u003e20\u003c/span\u003e] proposed novel fidelity metrics (FAED, FPCAD) that address shortcomings of aggregate metrics in capturing structural and class-conditional variation.\u003c/p\u003e\n\u003cp\u003e\u003cspan\u003e\u003c/span\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eB. GAN and VAE-Based Generative Models\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003c/p\u003e\n\u003cp\u003eGANs and VAEs remain dominant in deep generative mod- eling. Hernandez et al. [\u003cspan class=\"CitationRef\"\u003e1\u003c/span\u003e] reviewed GAN-based models in healthcare and noted their instability and poor mode coverage. Manousakas and Aydo\u0026uml;re [\u003cspan class=\"CitationRef\"\u003e4\u003c/span\u003e] benchmarked CTGAN, TVAE, and Gaussian Copula, highlighting that ML utility depends heavily on data complexity. Panagiotou et al. [\u003cspan class=\"CitationRef\"\u003e5\u003c/span\u003e] and Stoian et al. [\u003cspan class=\"CitationRef\"\u003e9\u003c/span\u003e] found that GAN-based generators often suffer from class imbalance and fairness distortions. CopulaGAN and CTGAN have shown promise in specific tabular tasks [\u003cspan class=\"CitationRef\"\u003e14\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e16\u003c/span\u003e], though performance varies significantly by dataset.\u003c/p\u003e\n\u003cp\u003e\u003cspan\u003e\u003c/span\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eC. Transformer and Diffusion-Based Methods\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003c/p\u003e\n\u003cp\u003eRecent innovations explore attention-based and diffusion models for tabular generation. Ma et al. [\u003cspan class=\"CitationRef\"\u003e13\u003c/span\u003e] introduced TabPFGen, a transformer-based model that performs well on small datasets. Kim et al. [\u003cspan class=\"CitationRef\"\u003e17\u003c/span\u003e] and Wang et al. [\u003cspan class=\"CitationRef\"\u003e19\u003c/span\u003e] surveyed these newer architectures, noting their expressive power but also higher training instability and reduced interpretability.\u003c/p\u003e\n\u003cp\u003e\u003cspan\u003e\u003c/span\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eD. Evaluation Metrics and ML Utility Assessment\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003c/p\u003e\n\u003cp\u003eSDMetrics is the most commonly used diagnostic suite [\u003cspan class=\"CitationRef\"\u003e7\u003c/span\u003e], offering shape, trend, and structure metrics. However, it lacks interpretability and fails to capture class-level fidelity breakdowns. Rashidi et al. [\u003cspan class=\"CitationRef\"\u003e6\u003c/span\u003e] combined SDMetrics with AutoML pipelines. Herurkar et al. [\u003cspan class=\"CitationRef\"\u003e20\u003c/span\u003e] argue that metrics like FAED better capture deviations. Other studies [\u003cspan class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e15\u003c/span\u003e] use Kolmogorov-Smirnov, TVD, and Spearman correlation to assess fidelity. ML utility is often tested using classifiers (SVM, LR, RF) and metrics like Accuracy, F1-score, and ROC AUC [\u003cspan class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e18\u003c/span\u003e]. However, these are sensitive to data imbalance and noise.\u003c/p\u003e\n\u003cp\u003e\u003cspan\u003e\u003c/span\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eE. Research Gaps\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003c/p\u003e\n\u003cp\u003eDespite growing progress, prior work still lacks:\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003e\n \u003cp\u003eInterpretable, feature-level diagnostics of fidelity;\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003eStrong correlation analysis between fidelity and ML util- ity.\u003c/p\u003e\n \u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eOur study addresses these gaps by benchmarking four pop- ular STDG models\u0026mdash;CTGAN, TVAE, Gaussian Copula, and CopulaGAN\u0026mdash;on a benchmark stroke prediction dataset. We introduce a two-phase generation and evaluation framework that combines SDMetrics, column-wise fidelity analysis using TVD, and downstream machine learning performance evalua- tion. Although both Random Forest and XGBoost classifiers were initially tested, Random Forest was retained for final analysis due to comparable performance.\u003c/p\u003e"},{"header":"III.\tMETHODOLOGY","content":"\u003cp\u003eThis section outlines the dataset, generative models, training setup, and evaluation framework used in this study. Our approach combines statistical and practical assessment metrics to evaluate synthetic tabular data fidelity and utility.\u003c/p\u003e\n\u003cp\u003e\u003cspan\u003e\u003c/span\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eA. Dataset Description\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003c/p\u003e\n\u003cp\u003eWe use a publicly available benchmark dataset for stroke prediction, originally based on the Kaggle dataset by Sori- ano [\u003cspan class=\"CitationRef\"\u003e29\u003c/span\u003e]. For modeling purposes, we utilized a synthetically expanded version of this dataset containing 15,000 samples and 22 features. The dataset includes a mix of numerical and categorical variables relevant to stroke risk, along with a binary target variable indicating stroke occurrence. After preprocessing, the final dataset used in this study consists of 8 numerical and 13 categorical features (including the target variable). The target classes are balanced in the dataset.\u003c/p\u003e\n\u003cp\u003e\u003cspan\u003e\u003c/span\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eB. Generative Models Evaluated\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003c/p\u003e\n\u003cp\u003eWe benchmark four popular generative models:\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003e\n \u003cp\u003e\u003cstrong\u003eCTGAN\u003c/strong\u003e \u0026ndash; A GAN-based tabular synthesizer that uses conditional vectors to handle categorical data.\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003e\u003cstrong\u003eTVAE\u003c/strong\u003e \u0026ndash; A VAE-based model designed for capturing nonlinear latent representations in tabular data.\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003e\u003cstrong\u003eGaussian Copula (GC)\u003c/strong\u003e \u0026ndash; A statistical model using fitted marginals and a Gaussian copula to capture inter-feature dependencies.\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003e\u003cstrong\u003eCopulaGAN\u003c/strong\u003e \u0026ndash; A hybrid model combining copula trans- formations with adversarial training.\u003c/p\u003e\n \u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003e\u003cspan\u003e\u003c/span\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eC. Two-Phase Training Strategy\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003c/p\u003e\n\u003cp\u003eWe use a two-phase generation framework:\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePhase 1: Epoch Tuning\u003c/strong\u003e \u0026ndash; Each deep model is trained on a 2,000-row subset. Multiple epoch settings are evaluated (200 to 500), and optimal settings are selected based on fidelity scores.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePhase 2: Full Generation\u003c/strong\u003e \u0026ndash; Final models are trained on the full 15,000-row dataset using the selected epochs. Each model generates three synthetic datasets of sizes 5K, 15K, and 25K. For GC and CopulaGAN, marginal distributions for numer- ical features are fitted using the fitter package. Best-fit distributions are selected based on minimizing the sum of squared errors.\u003c/p\u003e\n\u003cp\u003e\u003cspan\u003e\u003c/span\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eD. Evaluation Framework\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003c/p\u003e\n\u003cp\u003e\u003cspan\u003e\u003c/span\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e1. Column-Wise Fidelity Analysis\u003c/em\u003e: We propose a feature- level diagnostic approach based on class-conditional deviation. For a numerical feature \u003cem\u003eX\u003c/em\u003e with classes \u003cem\u003eC\u003c/em\u003e\u003csub\u003e1\u003c/sub\u003e and \u003cem\u003eC\u003c/em\u003e\u003csub\u003e2\u003c/sub\u003e, we compute the absolute difference between real and synthetic means per class:\u003c/p\u003e\n\u003cp\u003e\u003c/p\u003e\n\u003cp\u003e\u003cimg src=\"data:image/png;base64,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\"\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e3. Machine Learning Utility Evaluation:\u0026nbsp;\u003c/em\u003eTo assess practi- cal utility, we trained both a Random Forest and an XGBoost classifier\u0026nbsp;on\u0026nbsp;synthetic\u0026nbsp;data\u0026nbsp;and\u0026nbsp;evaluated\u0026nbsp;them\u0026nbsp;on\u0026nbsp;a\u0026nbsp;5,000-row real hold-out set. Since both models produced highly similar results across all metrics, we present Random Forest perfor- mance for clarity and consistency. Performance is measured using:\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003eSensitivity\u0026nbsp;(Recall\u0026nbsp;for\u0026nbsp;positive\u0026nbsp;class)\u003c/li\u003e\n \u003cli\u003eSpecificity\u0026nbsp;(Recall\u0026nbsp;for\u0026nbsp;negative\u0026nbsp;class)\u003c/li\u003e\n \u003cli\u003eBalanced Accuracy\u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003eROC AUC\u003c/p\u003e\n \u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eThis simulates real-world use cases where synthetic data is used to train ML models later applied to real environments.\u003c/p\u003e"},{"header":"IV.\tRESULTS","content":"\u003cp\u003eThis section presents fidelity and machine learning utility results for synthetic datasets generated using four generative models: CTGAN, TVAE, Gaussian Copula, and CopulaGAN. Evaluation includes SDMetrics scores, column-wise fidelity analysis, class distribution matching, and downstream classi- fication performance on real data.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eA. Epoch Tuning and Model Selection\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eEpoch tuning was performed for the three deep generative models (CTGAN, TVAE, and CopulaGAN) using a 2,000- row subset. SDMetrics scores and our custom column-wise diagnostic framework were used to select the optimal epoch. Table I summarizes the results.\u003c/p\u003e\n\u003cp\u003eTABLE I: Epoch Tuning Summary for Deep Generative Models\u003c/p\u003e\n\u003cp\u003e\u003cimg src=\"data:image/png;base64,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\"\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eB. Synthetic Data Generation and Selection\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eEach model was trained using its optimal epoch on the full 15,000-row real dataset. To ensure a fair comparison, we gen- erated 10 synthetic datasets of the same size (15,000 rows) for each model. The final dataset was selected using our column-wise diagnostic framework, prioritizing the highest number of Super and Moderate columns and the lowest number of Worst columns. In the case of ties, the dataset with the lowest average feature-level deviation was chosen.\u003c/p\u003e\n\u003cp\u003eTable II summarizes the column fidelity and average devi- ation for the selected dataset from each model.\u003c/p\u003e\n\u003cp\u003eTABLE II: Final Selected Synthetic Dataset Per Model (15K rows)\u003c/p\u003e\n\u003cp\u003e\u003cimg src=\"data:image/png;base64,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\"\u003e\u003c/p\u003e\n\u003cp\u003e\u003cspan\u003e\u0026nbsp;\u003c/span\u003eAs shown in Table II, the Gaussian Copula model achieved perfect fidelity labeling with all 21 features rated as Super and the lowest average feature deviation (0.0193), indicating strong statistical alignment with the real data. In contrast, TVAE exhibited higher deviation and more Worst-rated features, although it later demonstrated strong downstream classification performance.\u003c/p\u003e\n\u003cp\u003eTo further illustrate these results, Figure 2 shows the break- down of Super, Moderate, and Worst feature counts for each selected synthetic dataset. This visual comparison reinforces the sharp contrasts in column-wise fidelity across models.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eC. SDMetrics Fidelity Evaluation\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eWe\u0026nbsp;use\u0026nbsp;the\u0026nbsp;SDMetrics\u0026nbsp;library\u0026nbsp;from\u0026nbsp;the\u0026nbsp;SDV\u0026nbsp;framework to evaluate statistical fidelity of each synthetic dataset (5k, 15k, 25k) acorss all four models. Two official reports were computed:\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003e\u003cstrong\u003eDiagnostic Report\u003c/strong\u003e: Assesses schema compliance, null value constraints, and structural issues in the synthetic dataset.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eQuality Report\u003c/strong\u003e: Measures column-wise distributional similarity\u0026nbsp;and\u0026nbsp;pairwise\u0026nbsp;feature\u0026nbsp;relationships,\u0026nbsp;summarized as a Data Quality Score.\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eThe reports were computed using the run\u003cimg width=\"7\" height=\"1\" src=\"data:image/png;base64,R0lGODdhCgABAHcAACH+GlNvZnR3YXJlOiBNaWNyb3NvZnQgT2ZmaWNlACwAAAAACgABAIAAAAABAgMCA4SPBQA7\" alt=\"image\"\u003ediagnostic() and evaluate\u003cimg width=\"7\" height=\"1\" src=\"data:image/png;base64,R0lGODdhCgABAHcAACH+GlNvZnR3YXJlOiBNaWNyb3NvZnQgT2ZmaWNlACwAAAAACgABAIAAAAABAgMCA4SPBQA7\" alt=\"image\"\u003equality() functions respectively from the SDMetrics library. Table III summarizes the resulting scores only for 15k dataset\u0026nbsp;genereted\u0026nbsp;by\u0026nbsp;each\u0026nbsp;of\u0026nbsp;the\u0026nbsp;four\u0026nbsp;deep\u0026nbsp;generative\u0026nbsp;models.\u003c/p\u003e\n\u003cp\u003eTABLE III: SDMetrics Scores for Final Synthetic Datasets (15k rows)\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eModel\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 115px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDiagnostic\u0026nbsp;Score (%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 127px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eData\u0026nbsp;Quality\u0026nbsp;Score (%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eCTGAN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 115px;\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 127px;\"\u003e\n \u003cp\u003e85.44\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eTVAE\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 115px;\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 127px;\"\u003e\n \u003cp\u003e71.64\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eGaussian Copula\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 115px;\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 127px;\"\u003e\n \u003cp\u003e98.64\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eCopulaGAN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 115px;\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 127px;\"\u003e\n \u003cp\u003e86.50\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eAll models scored 100% in diagnostic validity, while the Gaussian\u0026nbsp;Copula\u0026nbsp;model\u0026nbsp;achieved\u0026nbsp;the\u0026nbsp;highest\u0026nbsp;overall\u0026nbsp;data\u0026nbsp;qual- ity. TVAE, despite a slightly lower quality score, performed better in downstream ML utility as we will see later.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eD. Target Class Distribution Comparison\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eTo assess whether generative models preserved the original outcome distribution, we analyzed the class proportions of the target variable \u003cem\u003eDiagnosis\u0026nbsp;\u003c/em\u003e(stroke vs. no-stroke) in each 15K-row\u0026nbsp;synthetic\u0026nbsp;dataset.\u0026nbsp;The\u0026nbsp;original\u0026nbsp;dataset\u0026nbsp;exhibited\u0026nbsp;near- perfect\u0026nbsp;balance\u0026nbsp;between\u0026nbsp;the\u0026nbsp;two\u0026nbsp;classes\u0026nbsp;(approximately\u0026nbsp;50.2% no-stroke and 49.8% stroke).\u003c/p\u003e\n\u003cp\u003eFigure 3 illustrates the distribution of class labels across the real dataset and synthetic datasets generated by CTGAN, TVAE, Gaussian Copula, and CopulaGAN.\u003c/p\u003e\n\u003cp\u003eWhile all models preserved the general structure of the binary class distribution, Gaussian Copula achieved the closest match to the original dataset (within 0.3%). TVAE over- represented the stroke class (by approximately 9%), while CT- GAN and CopulaGAN showed moderate under-representation of stroke cases. This alignment with class distribution is important for minimizing bias in downstream classification performance.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eE. Machine Learning Utility Evaluation\u003c/em\u003e:\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTo assess whether synthetic data supports downstream predictive tasks, we trained a Random Forest classifier on each synthetic dataset and evaluated it on a 5,000- row real hold-out set. The classifier was trained us- ing 10-fold stratified cross-validation with default hy- perparameters (n_estimators=100, max_depth=None, random_state=42).\u003c/p\u003e\n\u003cp\u003eTable IV reports the mean values across four evaluation metrics: Sensitivity, Specificity, Balanced Accuracy, and ROC AUC. For fair comparison, we include results for the real dataset at three sizes (5K, 15K, 25K) alongside synthetic datasets of equal sizes generated by each model.\u003c/p\u003e\n\u003cp\u003eTABLE IV: ML Utility Metrics (Mean over 10-Fold CV): RF Trained on Synthetic, Tested on Real\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eModel\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSize\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 52px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSensitivity\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 52px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSpecificity\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eBal. Accuracy\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eROC AUC\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003eOriginal (Real)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e5K\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 52px;\"\u003e\n \u003cp\u003e0.4402\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 52px;\"\u003e\n \u003cp\u003e0.5165\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e0.4784\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e0.4829\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e15K\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 52px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.4514\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 52px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.5356\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.4935\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.4925\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e25K\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 52px;\"\u003e\n \u003cp\u003e0.4587\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 52px;\"\u003e\n \u003cp\u003e0.5273\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e0.4930\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e0.4993\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003eTVAE\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e5K\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 52px;\"\u003e\n \u003cp\u003e0.8652\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 52px;\"\u003e\n \u003cp\u003e0.6768\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e0.7710\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e0.8613\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e15K\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 52px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.8781\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 52px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.7158\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.7970\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.8939\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e25K\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 52px;\"\u003e\n \u003cp\u003e0.8853\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 52px;\"\u003e\n \u003cp\u003e0.7283\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e0.8068\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e0.9012\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003eCTGAN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e5K\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 52px;\"\u003e\n \u003cp\u003e0.0548\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 52px;\"\u003e\n \u003cp\u003e0.9714\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e0.5131\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e0.5819\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e15K\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 52px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.0415\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 52px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.9831\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.5123\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.5864\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e25K\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 52px;\"\u003e\n \u003cp\u003e0.0640\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 52px;\"\u003e\n \u003cp\u003e0.9704\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e0.5172\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e0.5887\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003eGaussian Copula\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e5K\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 52px;\"\u003e\n \u003cp\u003e0.5642\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 52px;\"\u003e\n \u003cp\u003e0.4407\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e0.5024\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e0.5073\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e15K\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 52px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.4262\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 52px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.5730\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.4996\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.5016\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e25K\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 52px;\"\u003e\n \u003cp\u003e0.4756\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 52px;\"\u003e\n \u003cp\u003e0.5282\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e0.5019\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e0.5011\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003eCopulaGAN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e5K\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 52px;\"\u003e\n \u003cp\u003e0.0482\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 52px;\"\u003e\n \u003cp\u003e0.9769\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e0.5126\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e0.5798\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e15K\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 52px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.0539\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 52px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.9696\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.5118\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.5772\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e25K\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 52px;\"\u003e\n \u003cp\u003e0.0452\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 52px;\"\u003e\n \u003cp\u003e0.9768\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e0.5110\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 54px;\"\u003e\n \u003cp\u003e0.5853\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eAcross all models, TVAE demonstrated the strongest generalization ability, consistently outperforming other synthesizers and even the original dataset. Its performance improved with sample size, achieving the highest ROC AUC (0.9012) and balanced accuracy (0.8068) on the 25K dataset. This suggests strong preservation of class boundaries in the synthetic data.\u003c/p\u003e\n\u003cp\u003eCTGAN and CopulaGAN, while yielding high specificity, suffered from extremely low sensitivity. These results suggest overfitting\u0026nbsp;to\u0026nbsp;the\u0026nbsp;majority\u0026nbsp;class\u0026nbsp;and\u0026nbsp;confirm\u0026nbsp;observations\u0026nbsp;from earlier diagnostics\u0026mdash;particularly the imbalanced target class distribution.\u0026nbsp;Gaussian\u0026nbsp;Copula\u0026nbsp;achieved\u0026nbsp;moderate\u0026nbsp;performance, with relatively balanced sensitivity and specificity, aligning with its strong feature-wise fidelity but limited downstream utility.\u003c/p\u003e\n\u003cp\u003eFigure 4 provides a visual summary of the classification performance for 15K datasets across all models compared to the real data baseline.\u003c/p\u003e"},{"header":"V.\tDISCUSSION","content":"\u003cp\u003eOur evaluation reveals notable differences in how each synthetic data generation model balances fidelity and machine learning utility.\u003c/p\u003e\n\u003cp\u003eTVAE consistently delivered the strongest classification performance across all dataset sizes, achieving the highest ROC AUC and balanced accuracy. This suggests that, despite moderate\u0026nbsp;statistical\u0026nbsp;fidelity\u0026nbsp;in\u0026nbsp;terms\u0026nbsp;of\u0026nbsp;feature-level\u0026nbsp;deviations, TVAE\u0026nbsp;effectively\u0026nbsp;captures\u0026nbsp;class-separating\u0026nbsp;patterns\u0026nbsp;in\u0026nbsp;the\u0026nbsp;data. Its robustness to increasing dataset size further reinforces its utility for downstream predictive tasks.\u003c/p\u003e\n\u003cp\u003eGaussian Copula, on the other hand, demonstrated superior statistical fidelity\u0026mdash;achieving the highest number of Super columns and the lowest average feature-wise deviation. How- ever, its classification performance remained modest. This trade-off suggests that high fidelity in marginal and condi- tional distributions does not necessarily translate to improved learning of decision boundaries, particularly in nonlinear tasks. CTGAN and CopulaGAN, while structurally sound accord- ing to SDMetrics, exhibited poor classification utility. Both models suffered from extremely low sensitivity and skewed target class distributions, likely due to mode collapse and over- fitting to the majority class during adversarial training. Their failure to generalize downstream highlights the limitations of using aggregate fidelity scores alone to judge synthesis quality.\u003c/p\u003e\n\u003cp\u003eTo\u0026nbsp;make\u0026nbsp;this\u0026nbsp;contrast\u0026nbsp;more\u0026nbsp;explicit,\u0026nbsp;Table\u0026nbsp;V\u0026nbsp;summarizes\u0026nbsp;the relative ranking of each model based on fidelity and utility.\u003c/p\u003e\n\u003cp\u003eTABLE V: Model Ranking: Statistical Fidelity vs. ML Utility\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eModel\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eFidelity Rank\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eUtility Rank\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eTVAE\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e1\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eCTGAN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eGaussian Copula\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e1\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eCopulaGAN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 74px;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eAs seen in Table V, the highest fidelity score was achieved by Gaussian Copula, while TVAE dominated utility-based evaluation. The two deep generative models, CTGAN and CopulaGAN, lagged in both aspects. This divergence further reinforces the need to assess synthetic data not just through global similarity metrics, but through task-oriented validation aligned with intended applications.\u003c/p\u003e\n\u003cp\u003eOverall, our findings emphasize that fidelity and utility are not always aligned. Models like TVAE that may rank lower in statistical diagnostics can outperform others in real-world ML tasks. These results underline the importance of application- specific evaluation pipelines that combine both diagnostic and task-based assessments when benchmarking synthetic data generators.\u003c/p\u003e\n\u003cp\u003eThese findings motivate the need for more interpretable, utility-aware evaluation frameworks for synthetic data, which we further discuss in the next section.\u003c/p\u003e"},{"header":"VI. LIMITATIONS AND FUTURE WORK","content":"\u003cp\u003eWhile our evaluation framework provides detailed insights into fidelity and machine learning utility across multiple synthetic data generators, several limitations should be noted. First, the study focuses on a single healthcare dataset. While the results are indicative, generalizability to other domains (e.g., finance, marketing, social sciences) may be limited. Model behaviors\u0026mdash;particularly those involving marginal fi- delity and class preservation\u0026mdash;can vary across domains with\u003c/p\u003e\u003cp\u003edifferent statistical properties or class imbalances.\u003c/p\u003e\u003cp\u003eSecond, while our main downstream utility evaluation is reported using a Random Forest classifier due to its strong performance and interpretability, we also experimented with XGBoost and observed similar results. However, other models such as neural networks might yield different utility scores. Incorporating a broader set of classifiers or model-agnostic evaluation methods could provide a more comprehensive as- sessment of synthetic data utility.\u003c/p\u003e\u003cp\u003eThird, our fidelity evaluation focused on feature-wise class- conditional statistics and SDMetrics scores, which offer valu- able interpretability and standardized benchmarks. However, these metrics may not fully capture complex aspects such as\u003c/p\u003e\u003cp\u003ehigher-order correlations, fairness, privacy leakage, or tempo- ral consistency\u0026mdash;dimensions that are increasingly important in real-world, high-stakes applications.\u003c/p\u003e\u003cp\u003eMoreover, the selection of synthetic datasets was guided by column-level deviation thresholds to maintain transparency and tractability. While effective for identifying localized de- viations, this method may not account for global structural patterns or multi-feature interactions that could influence downstream model behavior.\u003c/p\u003e\u003cp\u003eFinally, the evaluation was done using 10-fold cross- validation on a fixed real hold-out set. Though this setup is reproducible, future work can explore multiple test sets, adaptive validation splits, or synthetic-to-synthetic evaluation loops to assess model robustness further.\u003c/p\u003e\u003cp\u003eIn future work, we aim to extend this framework to diverse datasets and classifiers, incorporate fairness and privacy met- rics, and develop a unified scoring system that aligns statistical fidelity with downstream utility in a domain-aware manner.\u003c/p\u003e"},{"header":"VII. CONCLUSION","content":"\u003cp\u003eThis study proposed a comprehensive and interpretable evaluation framework for assessing both statistical fidelity and machine learning utility of synthetic tabular data. Us- ing a benchmark dataset for stroke prediction, we evaluated four popular generative models\u0026mdash;TVAE, CTGAN, Gaussian Copula, and CopulaGAN\u0026mdash;across diagnostic, statistical, and downstream performance metrics.\u003c/p\u003e\u003cp\u003eOur results demonstrate that fidelity alone is not a re- liable indicator of utility. While Gaussian Copula achieved the highest statistical similarity to the original data, TVAE outperformed all other models in ML utility across sensitivity, specificity, balanced accuracy, and ROC AUC. Conversely, CT- GAN and CopulaGAN struggled to preserve target distribution and showed poor generalization.\u003c/p\u003e\u003cp\u003eThese findings highlight the need for application-specific evaluation pipelines that move beyond global scores and incor- porate column-level diagnostics and predictive performance. Our framework offers a practical approach for researchers and practitioners to select suitable synthetic data generators based on intended downstream tasks.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eA.K.D. and A.R. jointly conceived the research idea and designed the overall methodology of the study. A.K.D. led the development of the experimental pipeline, including model implementation, training, testing, and synthetic data generation. S.S. contributed to the evaluation and interpretation of results, particularly focusing on statistical diagnostics and machine learning utility analysis. A.K.D. also prepared the manuscript draft. All authors contributed to editing and revising the manuscript and approved the final version for submission.\u003c/p\u003e\u003ch2\u003eAcknowledgement\u003c/h2\u003e\u003cp\u003eThe authors would like to thank CHRIST (Deemed to be University), Bangalore, for providing institutional support throughout the research. We also extend our sincere gratitude to the Dr. Saravanan V (Scientist Grade -D), Indian Council of Medical Research (ICMR) for enabling this work as part of a student research internship.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eM. Hernandez, M. T. Perez, L. Serra, and L. J. Garcia, “Synthetic data generation for tabular health records: A systematic review,” \u003cem\u003eNeurocom-\u0026nbsp;puting\u003c/em\u003e, vol. 493, pp. 28–45, 2022.\u003c/li\u003e\n \u003cli\u003eA.\u0026nbsp;Figueira\u0026nbsp;and\u0026nbsp;B.\u0026nbsp;Vaz,\u0026nbsp;“Survey\u0026nbsp;on\u0026nbsp;synthetic\u0026nbsp;data\u0026nbsp;generation,\u0026nbsp;evaluation\u0026nbsp;methods and GANs,” \u003cem\u003eMathematics\u003c/em\u003e, vol. 10, no. 15, p. 2733, 2022.\u003c/li\u003e\n \u003cli\u003eE. Papadaki, A. G. Vrahatis, and S. Kotsiantis, “Exploring innovative\u0026nbsp;approaches to synthetic tabular data generation,” \u003cem\u003eElectronics\u003c/em\u003e, vol. 13,\u0026nbsp;no. 10, p. 1965, 2024.\u003c/li\u003e\n \u003cli\u003eD. Manousakas and S. Aydo¨re, “On the usefulness of synthetic tabular\u0026nbsp;data generation,” \u003cem\u003earXiv preprint arXiv:2306.15636\u003c/em\u003e, 2023.\u003c/li\u003e\n \u003cli\u003eE.\u0026nbsp;Panagiotou,\u0026nbsp;A.\u0026nbsp;Roy,\u0026nbsp;and\u0026nbsp;E.\u0026nbsp;Ntoutsi,\u0026nbsp;“Synthetic\u0026nbsp;tabular\u0026nbsp;data\u0026nbsp;generation\u0026nbsp;for class imbalance and fairness: A comparative study,” \u003cem\u003earXiv preprint\u0026nbsp;arXiv:2409.05215\u003c/em\u003e, 2024.\u003c/li\u003e\n \u003cli\u003eJ. Ma, H. Zhang, A. R. Mohamed, and M. Welling, “TabPFGen: Tabular data generation with TabPFN,” \u003cem\u003earXiv preprint arXiv:2406.05216\u003c/em\u003e, 2024.\u003c/li\u003e\n \u003cli\u003eH. H. Rashidi et al., “A novel and fully automated platform for synthetic tabular data generation and validation,” \u003cem\u003eScientific Reports\u003c/em\u003e, vol. 14, no. 1, p. 23312, 2024.\u003c/li\u003e\n \u003cli\u003eR. Shi et al., “A comprehensive survey of synthetic tabular data generation,” \u003cem\u003earXiv preprint arXiv:2504.16506\u003c/em\u003e, 2025.\u003c/li\u003e\n \u003cli\u003eM.\u0026nbsp;C.\u0026nbsp;Stoian\u0026nbsp;et\u0026nbsp;al.,\u0026nbsp;“How\u0026nbsp;realistic\u0026nbsp;is\u0026nbsp;your\u0026nbsp;synthetic\u0026nbsp;data?\u0026nbsp;Con-\u0026nbsp;straining deep generative models for tabular data,” \u003cem\u003earXiv preprint\u0026nbsp;arXiv:2402.04823\u003c/em\u003e, 2024.\u003c/li\u003e\n \u003cli\u003eD. Panfilo et al., “A deep learning-based pipeline for the generation of\u0026nbsp;synthetic tabular data,” \u003cem\u003eIEEE Access\u003c/em\u003e, vol. 11, pp. 63306–63323, 2023.\u003c/li\u003e\n \u003cli\u003eA. X. Wang et al., “Challenges and opportunities of generative models\u0026nbsp;on tabular data,” \u003cem\u003eApplied Soft Computing\u003c/em\u003e, vol. 141, p. 112223, 2024.\u003c/li\u003e\n \u003cli\u003eD.-K.\u0026nbsp;Kim,\u0026nbsp;Y.\u0026nbsp;Ryu,\u0026nbsp;H.\u0026nbsp;Lee,\u0026nbsp;and\u0026nbsp;H.\u0026nbsp;Choi,\u0026nbsp;“Generative\u0026nbsp;models\u0026nbsp;for\u0026nbsp;tabular\u0026nbsp;data:\u0026nbsp;A\u0026nbsp;review,”\u0026nbsp;\u003cem\u003eJournal\u0026nbsp;of\u0026nbsp;Mechanical\u0026nbsp;Science\u0026nbsp;and\u0026nbsp;Technology\u003c/em\u003e,\u0026nbsp;vol.\u0026nbsp;38,\u0026nbsp;no. 9, pp. 4989–5005, 2024.\u003c/li\u003e\n \u003cli\u003eC. Hassan, R. Salomone, and K. Mengersen, “Deep generative mod-\u0026nbsp;els, synthetic tabular data, and differential privacy: An overview and\u0026nbsp;synthesis,” \u003cem\u003earXiv preprint arXiv:2307.15424\u003c/em\u003e, 2023.\u003c/li\u003e\n \u003cli\u003eT.\u0026nbsp;Liu\u0026nbsp;et\u0026nbsp;al.,\u0026nbsp;“Goggle:\u0026nbsp;Generative\u0026nbsp;modelling\u0026nbsp;for\u0026nbsp;tabular\u0026nbsp;data\u0026nbsp;by\u0026nbsp;learning\u0026nbsp;relational structure,” in \u003cem\u003eProc. Int. Conf. Learn. Representations (ICLR)\u003c/em\u003e,\u0026nbsp;2023.\u003c/li\u003e\n \u003cli\u003eL. Xu and K. Veeramachaneni, “Synthesizing tabular data using gener-\u0026nbsp;ative adversarial networks,” \u003cem\u003earXiv preprint arXiv:1811.11264\u003c/em\u003e, 2018.\u003c/li\u003e\n \u003cli\u003eK. Liu and R. B. Altman, “Conditional generative models for synthetic\u0026nbsp;tabular data: Applications for precision medicine and diverse represen-\u0026nbsp;tations,” \u003cem\u003eAnnu. Rev. Biomed. Data Sci.\u003c/em\u003e, vol. 8, 2025.\u003c/li\u003e\n \u003cli\u003eB.\u0026nbsp;Wen\u0026nbsp;et\u0026nbsp;al.,\u0026nbsp;“Causal-TGAN:\u0026nbsp;Generating\u0026nbsp;tabular\u0026nbsp;data\u0026nbsp;using\u0026nbsp;causal\u0026nbsp;gen-\u0026nbsp;erative adversarial networks,” \u003cem\u003earXiv preprint arXiv:2104.10680\u003c/em\u003e, 2021.\u003c/li\u003e\n \u003cli\u003eA. X. Wang, B. P. Nguyen, and C. R. Simpson, “Generative AI for\u0026nbsp;tabular\u0026nbsp;data\u0026nbsp;synthesis,”\u0026nbsp;in\u0026nbsp;\u003cem\u003eProc.\u0026nbsp;Pacific-Asia\u0026nbsp;Conf.\u0026nbsp;Knowl.\u0026nbsp;Discov.\u0026nbsp;Data\u0026nbsp;Mining (PAKDD)\u003c/em\u003e, Springer, 2025.\u003c/li\u003e\n \u003cli\u003eX.\u0026nbsp;Wen\u0026nbsp;et\u0026nbsp;al.,\u0026nbsp;“From\u0026nbsp;supervised\u0026nbsp;to\u0026nbsp;generative:\u0026nbsp;A\u0026nbsp;novel\u0026nbsp;paradigm\u0026nbsp;for tabular deep learning with large language models,” in \u003cem\u003eProc. ACM\u0026nbsp;SIGKDD Conf. Knowl. Discov. Data Mining\u003c/em\u003e, 2024.\u003c/li\u003e\n \u003cli\u003eD.\u0026nbsp;Herurkar,\u0026nbsp;A.\u0026nbsp;Ali,\u0026nbsp;and\u0026nbsp;A.\u0026nbsp;Dengel,\u0026nbsp;“Evaluating\u0026nbsp;generative\u0026nbsp;models\u0026nbsp;for tabular data: Novel metrics and benchmarking,” \u003cem\u003earXiv preprint\u0026nbsp;arXiv:2504.20900\u003c/em\u003e, 2025.\u003c/li\u003e\n \u003cli\u003eA. Gonzales, G. Guruswamy, and S. R. Smith, “Synthetic data in health\u0026nbsp;care:\u0026nbsp;A\u0026nbsp;narrative\u0026nbsp;review,”\u0026nbsp;\u003cem\u003ePLOS\u0026nbsp;Digital\u0026nbsp;Health\u003c/em\u003e,\u0026nbsp;vol.\u0026nbsp;2,\u0026nbsp;no.\u0026nbsp;1,\u0026nbsp;e0000082,\u0026nbsp;2023.\u003c/li\u003e\n \u003cli\u003eP.\u0026nbsp;Davis,\u0026nbsp;R.\u0026nbsp;Lay-Yee,\u0026nbsp;and\u0026nbsp;J.\u0026nbsp;Pearson,\u0026nbsp;“Using\u0026nbsp;micro-simulation\u0026nbsp;to\u0026nbsp;create\u0026nbsp;a synthesised data set and test policy options: The case of health service\u0026nbsp;effects under demographic ageing,” \u003cem\u003eHealth Policy\u003c/em\u003e, vol. 97, no. 2–3, pp.\u0026nbsp;267–274, 2010.\u003c/li\u003e\n \u003cli\u003eC. Ngufor, H. Van, B. S. Caffo, N. D. Shah, and R. G. McCoy, “Mixed\u0026nbsp;effect\u0026nbsp;machine\u0026nbsp;learning:\u0026nbsp;A\u0026nbsp;framework\u0026nbsp;for\u0026nbsp;predicting\u0026nbsp;longitudinal\u0026nbsp;change\u0026nbsp;in hemoglobin A1c,” \u003cem\u003eJournal of Biomedical Informatics\u003c/em\u003e, vol. 89, pp.\u0026nbsp;56–67, 2019.\u003c/li\u003e\n \u003cli\u003eW. T. Enanoria et al., “The effect of contact investigations and public\u0026nbsp;health interventions in the control and prevention of measles transmis-\u0026nbsp;sion: A simulation study,” \u003cem\u003ePLOS ONE\u003c/em\u003e, vol. 11, no. 12, e0167160, 2016.\u003c/li\u003e\n \u003cli\u003eT. Laderas et al., “Teaching data science fundamentals through realistic\u0026nbsp;synthetic clinical cardiovascular data,” \u003cem\u003ebioRxiv\u003c/em\u003e, preprint 232611, 2017.\u003c/li\u003e\n \u003cli\u003eSMART Health IT, “SMART on FHIR developer sandbox documenta-\u0026nbsp;tion,” 2017.\u003c/li\u003e\n \u003cli\u003eCenters for Disease Control and Prevention (CDC), “Public-use linked\u0026nbsp;mortality file: 2015 linked mortality file description,” Mar. 2020.\u003c/li\u003e\n \u003cli\u003eSDV\u0026nbsp;Documentation,\u0026nbsp;“Single\u0026nbsp;Table\u0026nbsp;Synthesizers,”\u0026nbsp;SDV\u0026nbsp;Dev\u0026nbsp;Documen-\u0026nbsp;tation. [Online]. Available: https://docs.sdv.dev/sdv\u003c/li\u003e\n \u003cli\u003eF. Soriano, “Stroke Prediction Dataset,” Kaggle, 2021. [Online]. Available: https://www.kaggle.com/datasets/fedesoriano/ stroke-prediction-dataset\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"cluster-computing","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"Learn more about [Cluster Computing](https://www.springer.com/journal/10586)","snPcode":"10586","submissionUrl":"https://submission.nature.com/new-submission/10586/3","title":"Cluster Computing","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Synthetic Data, Tabular Data Generation, CT- GAN, TVAE, Gaussian Copula, Machine Learning Evaluation, Data Fidelity","lastPublishedDoi":"10.21203/rs.3.rs-7287372/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7287372/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eSynthetic tabular data offers a promising solution for enabling privacy-preserving machine learning in sensitive domains such as healthcare. However, assessing the fidelity and utility of such data remains challenging. In this study, we evaluate four generative models\u0026mdash;CTGAN, TVAE, Gaussian Copula, and CopulaGAN\u0026mdash;on a benchmark dataset for stroke prediction. We propose a two-phase generation and evaluation framework that combines statistical diagnostics with feature-level fidelity analysis and downstream classification performance. Our findings highlight significant variation across models, with TVAE and Gaussian Copula achieving superior fidelity and generalization. The results demonstrate that high structural similarity does not always guarantee practical machine learning utility.\u003c/p\u003e","manuscriptTitle":"Evaluating Fidelity and Machine Learning Utility of Synthetic Tabular Data Generated Using Generative Models","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-09-17 13:41:15","doi":"10.21203/rs.3.rs-7287372/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"editorInvitedReview","content":"","date":"2026-05-05T23:29:26+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"145482980954007588186232639198593157100","date":"2026-05-04T00:47:21+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"101959305593984138368632119831233859401","date":"2026-05-03T15:42:22+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"23264700169985611687472971136760300072","date":"2025-09-10T10:50:06+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-09-10T06:33:58+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-09-10T06:22:37+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-08-05T08:59:50+00:00","index":"","fulltext":""},{"type":"submitted","content":"Cluster Computing","date":"2025-08-04T06:00:50+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"cluster-computing","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"Learn more about [Cluster Computing](https://www.springer.com/journal/10586)","snPcode":"10586","submissionUrl":"https://submission.nature.com/new-submission/10586/3","title":"Cluster Computing","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"79627237-2087-4868-90fe-cbde72cd349f","owner":[],"postedDate":"September 17th, 2025","published":true,"recentEditorialEvents":[{"type":"editorInvitedReview","content":"","date":"2026-05-05T23:29:26+00:00","index":80,"fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2025-09-17T13:41:15+00:00","versionOfRecord":[],"versionCreatedAt":"2025-09-17 13:41:15","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7287372","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7287372","identity":"rs-7287372","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}