DED-AM-MMC dataset

Dataset Description: Multi-Defect Feature Dataset for SHAP-Based Analysis in Directed Energy Deposition Additive Manufacturing This dataset contains feature representations extracted from a fine-tuned VideoMAE model applied to melt pool monitoring videos captured using CLAMIR and FLIR mid-wave infrared (MWIR) cameras during a directed energy deposition (DED) additive manufacturing (AM) process. The data supports explainable machine learning workflows for classifying defect severity levels in additively manufactured Ni-WC metal matrix composites. The dataset includes six defect categories: GP – Gas Porosity SP – Shrinkage Porosity EDB – Excessive Dilution with Base Material NCD – Non-uniform Carbide Distribution ECD – Excessive Carbide Dissolution RPHP – Reprecipitated Hard Phases Each defect category is labeled with three classes representing defect extent: Extent-free (absence) Low extent High extent Each of the six categories includes two types of feature datasets (CLAMIR-based and FLIR-based), giving a total of 12 CSV files. To facilitate usability: The six feature selection subsets (best_subset_gp_*) are grouped and provided in a single archive: feature_indices_files.zip. 'gp' in this example refers to Gas Porosity and similarly 'sp', 'edb', 'ncd', 'ecd', and 'rphp' are used in the file nomenclature to refer to defect categories. The classes are reflected by their index '0', '1', and '2' referring to absent, low and high respectively. The six training and test sets (train_features_gp_* and test_features_gp_*) are archived as train_test_files.zip. Test csv files for CLAMIR and FLIR are separated into stable and unstable version wheras the noisy version was genereted by processing the stable csv files during modeling. This dataset supports research in SHAP-based explainable AI, feature selection, sensor fusion, and multi-class defect classification in DED AM.