Artificial Intelligence (AI) models for melanoma detection have reported high accuracy. We retrospectively assessed performance of lesion malignancy risk and change detection models to detect melanoma. A total of n=2286 non-excised naevi and n=37 excised lesions (incl. n=13 melanomas) in n=13 patients were analyzed. The median lesion change varied between individuals (0.0-0.7), where higher scores denote greater change. The median change score in excised lesions was 1.5 (IQR: 0.0-4.2). Median change scores for melanomas and benign lesions were 4.2 (IQR: 2.2-5.9) and 0.0 (IQR: 0.0-1.6), respectively (p=0.036). Sensitivity and specificity of the malignancy risk assessment of excised lesions were 0.85 (95% CI: 0.55, 0.89) and 0.62 (95% CI: 0.41, 0.81), respectively. Median malignancy risk scores for melanomas and benign lesions were 7.7/10 and 0.5/10, respectively (p=0.004). Combining AI models and adjusting for individual-specific change improved melanoma detection and reduced the number of benign lesion excisions. Specifically, the combined AI approach recommended either excision or clinical review of melanomas. Three benign lesions would have been recommended for excision and three for clinical review, while the excision of eight would have been prevented. Integrating both AI models in clinical practice offers complementary information which may support clinicians in the detection of melanoma.