ahmad-fakhar/skin_cancer

Skin Cancer Classification — ConvNeXt-Base + DQN

Architecture: ConvNeXt-Base backbone + DQN Q-head (7 classes)
Dataset: HAM10000 — 7-class dermoscopic image classification
Split: By lesion_id (data-leakage free)

Performance

• Test Accuracy: 42.28%
• Weighted F1: 0.4493
• Macro F1: 0.3736
• Macro ROC-AUC: 0.8565

Classes

{ "akiec": 0, "bcc": 1, "bkl": 2, "df": 3, "mel": 4, "nv": 5, "vasc": 6 }