I built a melanoma detection workflow using the ISIC dataset: lesion segmentation with U‑Net, handcrafted and embedded features aligned with ABCDE rules, and classification with scikit‑learn (Random Forest classification).

Focus areas :

preprocessing, segmentation quality, and feature‑based interpretability for decision support.

Technical aspect :

U‑Net segmentation, color/shape/texture descriptors, classical ML classifiers.

Results :

Improved lesion isolation and consistent feature extraction for downstream classification. I achieved a balanced accuracy of 0.544 with Random Forests on handcrafted features which rank me in the top 25% of challenge participants.

You can find the code on my github repository : Here