Additively manufactured auxetic structures are lightweight, durable, and highly adaptable, with a negative Poisson’s ratio enabling tailored performance. Unlike conventional materials, their mechanical behavior is dictated by design rather than material composition, making them ideal for applications in engineering, impact protection and biomechanics [1]. Energy absorption capability is a key feature of auxetic structures, enhancing their efficiency under impact and dynamic loading conditions. This study aims to optimize energy absorption of a previously proposed auxetic unit cell [2] using an interpretable machine learning (ML) approach. To develop a reliable predictive model, a high-quality dataset is generated through experimentally validated finite element analysis (FEA). First, uniaxial compression tests are conducted on thermoplastic polyurethane (TPU) cylinders in accordance with polymeric materials standard to characterize the material properties. Then, a set of 320 auxetic unit cells with varying mesoscale geometrical is simulated to evaluate their energy absorption under compression. These data served as input for training a feed-forward multi layer perceptron model, consisting of an input layer corresponding to the geometrical parameters, hidden layers and neurons respectively utilizing activation functions, and an output layer predicting energy absorption values. The model was trained using the back-propagation algorithm, optimizing the mean squared error loss function. Which achieves high prediction accuracy on unseen test data. Finally, to validate the ML-driven predictions, the optimal unit cell identified by the model is fabricated via fused filament fabrication (FFF) and experimentally tested against randomly designed counterparts. The results confirm the effectiveness of ML in guiding auxetic structures design, offering a robust methodology for optimizing energy absorbing structures in advanced engineering applications.