<p>This study introduces a hybrid detection–prediction framework for quantifying the number of defects formed under different SLM processing conditions in Hastelloy X alloy. Microscopic defects in additively manufactured parts produced via the selective laser melting process (SLM) were analyzed by capturing cross-sectional images and automatically detecting defects. Key process variables—scan speed, laser power, and hatch distance—were correlated with the detected defects to predict their influence on the process. A convolutional neural network based on the YOLO (you only look once) architecture was trained on microscopic images to automatically locate and classify three primary defect types: crack, lack of fusion (LOF), and porosity. The detected defect count was subsequently correlated with key process parameters using several machine learning models. Among the evaluated models, the Random Forest approach demonstrated the best predictive performance with the lowest prediction error. The analysis revealed that scan speed and laser power are the most influential parameters governing defect formation, while hatch distance has a comparatively smaller effect. Statistical analysis of the volumetric energy density (VED) further showed that LOF and porosity defects exhibit significant negative correlations with VED, indicating that higher energy input improves melt pool stability and consolidation. The model predictions indicate that a laser power of approximately 230 W, a scan speed of 1200 mm/s, and a hatch distance near 0.07 mm minimize crack and LOF defects.</p>