This study addresses AI bias in classifying insect bites on darker skin tones. Using 879 dermatologist-verified images categorized by Fitzpatrick Skin Type (FST), a CNN (56.8% accuracy) and InceptionV3 (79.5% accuracy) were tested, revealing a 21–33% accuracy drop on darker tones (FST III–VI). To improve performance, pretraining on the Stanford DDI dataset (656 images) and task arithmetic were applied. Both methods showed potential in enhancing accuracy through transfer learning and model editing.