This paper presents a multi-objective optimization study of thin-walled prismatic frusta (truncated pyramids) designed as crash energy absorbers. The research involved over 130 geometric variants, including both hollow and EPP foam-filled configurations, analyzed using finite element simulations and validated through experimental impact tests. A data-driven optimization algorithm, based on artificial neural networks (ANNs), was developed to enable fast prediction of crashworthiness parameters such as absorbed energy, peak crushing force, and total efficiency. To demonstrate practical applicability, the algorithm was used to optimize structural elements of a homologation vehicle frame in accordance with Regulation No. 29 of the UN ECE, achieving targeted energy absorption while respecting design constraints. Increasing the apex angle α reduces Peak Crushing Force (PCF) in both hollow and filled frusta. Foam filling increases PCF, but its effect on Crush Load Efficiency (CLE) varies with α. Filling improves Total Efficiency (TE) in hollow frusta significantly, while the effect is geometry-dependent in filled variants. The ANN-based tool reduces design iteration time and supports early-stage decisions in the development of crashworthy structures. These results contribute to the advancement of lightweight, energy-efficient safety components in vehicle design.