Accurate characterisation of blood dielectric properties is essential for data-driven biomedical sensing, yet experimental datasets are often limited to a few discrete hemoglobin (Hb) concentrations. This constraint hinders the development of robust data-driven models. To address this, the present study introduces a framework for generating synthetic blood permittivity spectra from sparse measurements. Four data-generation strategies were investigated, combining interpolation-based techniques and probabilistic models to extend Hb-dependent spectral coverage across the measured frequency range. Model performance was evaluated using Earth Mover’s Distance (EMD) for spectral similarity, Cole–Cole parameter analysis for physical consistency, variance preservation metrics, and Hb prediction using XGBoost. The results indicate that interpolation-based approaches achieve the highest reconstruction accuracy, while Conditional Bayesian principal component analysis (Conditional BPCA) produces smooth and physically consistent spectra with stable variability characteristics. Across all methods, the generated datasets maintained sufficient Hb-related information to support reliable prediction. These findings demonstrate that the proposed framework enables effective expansion of limited dielectric datasets while supporting a multi-criteria evaluation of synthetic data quality, including fidelity, variability, and predictive relevance.