Image Classifier — Non-Destructive Identification System (end-to-end training/prediction workflow)
Built an image-classification ML workflow where users upload training data and define regions of interest to train an ensemble model. Ran batch predictions through a browser interface and evaluated performance using leave-one-image-out cross-validation. Added interpretability visuals such as confusion matrices, PCA, and spectral plots to understand labeling/prediction quality. • Defined regions of interest for training inputs • Trained soft-voting ensemble (SVM, Random Forest, XGBoost) • Performed leave-one-image-out cross-validation • Produced interpretability visualizations for evaluation