Two-dimensional (2D) halide perovskites are forthcoming as promising candidates for next-generation ferroelectric and optoelectronic devices. Traditional discovery methods are computationally intensive and slow because they cover a large memory and space. In this study, we used a graph neural network (GNN) framework to more quickly forecast the structural and ferroelectric properties of 2D halide perovskites. Density functional theory (DFT) was used to generate a diverse dataset of polar and non-polar perovskite monolayers, including formation energies, spontaneous polarization, and band gaps. A GNN model that could precisely replicate DFT-level property trends at a fraction of the cost was trained using these DFT-relaxed structures recorded as crystal graphs. To facilitate rapid structural relaxation, a GNN-based interatomic potential was also developed, which identified key polar distortions and showed good agreement with DFT geometries. Some 2D halide perovskites with expected significant out-of-plane polarization and excellent stability are revealed by high-throughput screening throughout a wide A–B–X chemical space.