Electroacoustic absorbers offer a promising strategy for noise control systems in applications where absorber thickness is a limiting factor, such as low-frequency damping in room acoustics. These devices consist of an enclosed cavity with an electrodynamic loudspeaker, whose acoustic impedance can be adjusted. Traditionally, these devices have been modeled as single-degree-of-freedom (SDOF) systems [1], and they have demonstrated good acoustic absorption results near resonance, particularly in active noise control strategies [2]. However, due to uncertainties in model parameters (e.g., errors or changes in Thiele-Small parameters), model order (neglected higher-order modes), and other acoustic characteristics (cavity modes), the behavior of these devices often deviates from model predictions at both low and high frequencies. Attempts have been made to represent these electroacoustic systems using multiple-degree-of-freedom (MDOF) models [3]. While these models perform well in narrow-band operation, they lack reliability in predicting system dynamics over large frequency ranges. This poses a challenge for designing control laws based on model predictive control (MPC) systems or generating data for training machine learning control algorithms. MPC algorithms require models that accurately represent the system under control, while machine learning algorithms benefit from training data that closely reflects the target application. To address these limitations and account for model uncertainties, this work proposes a data-driven strategy based on an autoencoder [4] to identify system behavior across low and high frequencies and incorporate uncertainties into the simulated models. A single-degree-of-freedom model is used to represent the theoretical behavior of the electroacoustic absorber in scenarios with varying mass, stiffness, and damping coefficients. The system is tested in multiple configurations, both numerically and experimentally, and its acoustic impedance is evaluated. Then, an autoencoder is trained using the simulated acoustic impedance as input and experimental measurements as output, mapping the experimental uncertainty to the numerical model predictions. The proposed approach is evaluated on unseen experimental data to assess the generalization capabilities of the autoencoder strategy.