Supercontinuum (SC) generation, driven by intricate nonlinear effects and dispersion dynamics, is essential in spectroscopy, optical coherence tomography, and high-resolution imaging. Conventional numerical modeling, while accurate, is computationally intensive, limiting scalability. We present a multilayer perceptron (MLP)-based deep learning approach to predict SC spectra from key input parameters. To train the model, we prepared a unique dataset of 1,944 samples based on SiC rib waveguides with two cladding materials—SiO2 and MgF2. The dataset is highly expandable, allowing broader generalization across different materials and waveguide structures. Achieving a mean squared error (MSE) of 0.0022, the model delivers high-accuracy SC prediction while drastically reducing computation time. This framework provides a fast, accurate, and scalable tool for SC modeling and the design of nonlinear photonic systems.