Metasurfaces comprised of sub-wavelength structures, possess remarkable electromagnetic (EM) wave manipulation capabilities. Their application as radar absorbers has gained widespread recognition, particularly in modern stealth technology, where their main role is to minimize the radar cross-section (RCS) of military assets. Conventional radar absorber design is tedious because of its time-consuming, computationally intensive, iterative nature, and demand for a high level of expertise. In contrast, the emergence of machine/deep learning-based metasurface design for RCS reduction represents a rapidly evolving field. This approach offers automated and computationally efficient means to generate radar absorber designs. In this article, an inverse approach, using machine/deep learning methodology is presented for multilayered broadband microwave absorber. The proposed method is primarily based on geometry and absorption characteristics. The proposed design is based on an in-depth understanding of the behavior of an optimized, practically implementable impedance sheet-based meta-atom, and its electromagnetic variations relative to its overall dimensions and thickness. The meta-atom selected for this modelling is a Jerusalem cross and the data set used for the model consists of the geometry, thickness, and corresponding absorptivity. The predicted results of machine and deep learning models are further validated by simulating EM performance using full wave simulation software. The results predicted by the models are in good agreement with EM simulations from the C to K bands of frequencies. This model can be employed to create radar-absorbing structures using a single meta-atom design, tailored to various frequency bands.