Machine Learning (ML) is reshaping the landscape of materials design by enabling predictive and generative capabilities that go beyond conventional heuristics. This chapter presents a comprehensive synthesis of state-of-the-art techniques in forward modeling, inverse design, and representation learning for materials science, highlighting the interplay between data-driven inference and domain-specific physical constraints. Beginning with the foundations of supervised property prediction, the chapter explores advanced generative models, graph-based neural architectures, and the integration of physics priors to ensure model validity and generalizability. Through evaluations of model robustness, domain transferability, and real-world deployment barriers, it exposes both the opportunities and limitations of current methods. The discussion culminates in future directions such as federated collaboration, regulatory-aware explainability, and open source benchmarking to build resilient and transparent pipelines. This work provides a structured roadmap for realizing scalable, reliable, and interpretable ML systems for next-generation materials innovation.