<p>Laser Directed Energy Deposition (LP-DED) is a prominent laser additive manufacturing process and has been increasingly adopted in aerospace, energy, and other high-end manufacturing fields, particularly for repairing and producing large, geometrically complex components. However, LP-DED involves intense thermal input and strongly coupled interactions among heat transfer, melt flow, mass transport, and solidification, leading to pronounced non-stationary behavior. Consequently, deposition outcomes are highly sensitive to disturbances and parameter fluctuations, and typical defects—including porosity, cracking, and distortion—may occur, undermining geometric accuracy and part-to-part consistency. Enhancing manufacturing reliability therefore calls for a systematic understanding of defect evolution as well as timely, data-driven defect identification. In this review, we first summarize major defect categories in LP-DED, their formation routes, and dominant contributing factors. We then survey recent machine learning-enabled approaches for defect recognition and process modeling, with emphasis on multisource monitoring signals such as optical imaging, thermal data, acoustic emission, and spectroscopic measurements. Given the limited number of studies specifically targeting LP-DED, representative research from Selective Laser Melting (SLM) and Laser Powder Bed Fusion (LPBF) is also included to broaden methodological perspectives and provide complementary evidence. To reduce the heavy reliance of supervised learning on labeled data, we further discuss the potential of semi-supervised, unsupervised, and multi-task learning for LP-DED under realistic data-scarce settings. Finally, future directions are outlined toward building robust multisource sensing and data analytics pipelines, improving model generalization across materials and geometries, and advancing practical, reliable defect monitoring for high-quality LP-DED manufacturing.</p>