<p>Halide perovskites combine exceptional optoelectronic properties with compositional flexibility, but conventional discovery workflows are slow and resource-intensive. Machine learning (ML) now accelerates this cycle by learning structure–property relationships from experimental and density functional theory (DFT) data to predict stability, band gap (Eg), defect tolerance, and synthesis windows. This review surveys recent ML advances across hybrid organic–inorganic and double-perovskite families, including descriptors and feature engineering, classification/regression models, and uncertainty-aware screening. We summarize models that identify stable chemistries, prioritize candidates with target Eg, and guide synthesis, and we outline best practices for data curation and model validation. Remaining challenges include small/biased datasets, limited interpretability, and transfer to out-of-distribution chemistries. We close with an outlook on integrating ML with automated experiments and high-throughput computation to enable faster, sustainable perovskite discovery.</p>