Melanoma Skin Cancer Detection Using Deep Learning Algorithm

Abstract

Melanoma is one of the most aggressive forms of skin cancer, and early detection plays a vital role in reducing mortality rates. Traditional diagnostic methods rely heavily on expert dermatologists and biopsy confirmation, which can be time-consuming, subjective, and resource-intensive. With advances in artificial intelligence, deep learning has emerged as a powerful approach for medical image classification. This paper presents a convolutional neural network (CNN)-based model for melanoma detection using the HAM10000 dataset. The dataset of 10,015 dermoscopic images was pre-processed through normalization, resizing, and augmentation to address class imbalance. The proposed CNN model was trained and evaluated using multiple metrics, achieving 97% accuracy, 85% precision, 79% recall, and 82% F1-score. Class-wise performance and graphical analysis, including training curves and a confusion matrix, validated the robustness of the model. The study demonstrates that CNNs can serve as effective tools for melanoma detection, offering potential clinical support for dermatologists. Future work should focus on improving sensitivity, addressing dataset imbalance, and integrating explainable AI techniques to enhance clinical adoption.

Introduction

Skin cancer is a growing global health concern, with melanoma being the most aggressive and deadly form due to its high metastatic potential. Although non-melanoma skin cancers are more common, melanoma accounts for a disproportionate number of deaths. Early detection is critical, as survival exceeds 95% when diagnosed early but drops sharply once metastasis occurs. Traditional diagnosis relies on expert visual examination and biopsies, which face challenges such as limited access to dermatologists, variability in diagnosis, and delays.

Advances in artificial intelligence, particularly deep learning and convolutional neural networks (CNNs), have created new opportunities for automated melanoma detection from dermoscopic images. This study proposes a CNN-based framework using the HAM10000 dataset to classify seven types of skin lesions, with an emphasis on improving early melanoma detection. The methodology includes extensive image pre-processing, data augmentation, a custom CNN architecture, and evaluation using multiple performance metrics.

The proposed model achieved an overall accuracy of 97%, with strong precision, recall, and F1-score values, demonstrating effective classification across lesion types. Melanoma detection showed high precision but slightly lower recall, reflecting the clinical difficulty of distinguishing melanoma from visually similar benign lesions. Training and validation results indicated good generalization without overfitting.

The findings confirm that CNN-based deep learning models can serve as reliable supportive tools for dermatologists, improving diagnostic consistency and access to care, especially in resource-limited settings. While not a replacement for clinical expertise, such AI systems have strong potential to reduce misdiagnosis, enable early detection, and ultimately save lives, provided further optimization, explainability, and clinical validation are pursued.

Conclusion

The study presented in this dissertation investigated the application of deep learning, specifically convolutional neural networks (CNNs), for the detection and classification of melanoma skin cancer using the HAM10000 dataset. The work addressed one of the most significant challenges in dermatology: the early and accurate detection of melanoma, which remains a leading cause of skin cancer-related deaths worldwide. By systematically implementing a robust methodology that included pre-processing, augmentation, model design, training, and performance evaluation, the research demonstrated the power of AI in improving diagnostic accuracy and supporting dermatologists in clinical practice.The proposed CNN model achieved 97% overall accuracy, supported by a precision of 85%, recall of 79%, and an F1-score of 82%. These results underscore the model’s strong generalization ability across diverse lesion categories, while also highlighting the critical challenge of improving sensitivity for melanoma cases. The analysis revealed that the model performed exceptionally well in detecting majority classes such as melanocytic nevi but exhibited slightly reduced performance in minority classes, including melanoma and actinic keratoses. This observation reflects the underlying dataset imbalance and suggests areas for future improvement.Despite these limitations, the research contributes valuable insights to the growing field of AI in dermatology. It establishes that CNN-based architectures can deliver exceptional accuracy in melanoma detection and opens avenues for more advanced research to enhance performance, scalability, and clinical integration.

References

[1] Kirar, B. S., Agrawal, D. K., Baghel, R. K., & Kirar, S. (2017). Glaucoma image analysis using discrete wavelet transform. Journal of Engineering, Science and Management Education, 10(2), 114–118. https://www.indianjournals.com/ijor.aspx?target=ijor:jesme&volume=10&issue=2&article=008 [2] Kirar, B. S., Agrawal, D. K., & Kirar, S. (2020). Automated glaucoma detection using variational mode decomposition from fundus images. Indian Journal of Public Health Research & Development, 11(6), 1146–1153. https://doi.org/10.37506/ijphrd.v11i6.9974 [3] Kirar, B. S., Agrawal, D. K., & Kirar, S. (2022). Glaucoma detection using image channels and discrete wavelet transform. IETE Journal of Research, 68(6), 4421–4428. https://doi.org/10.1080/03772063.2021.1985390 [4] Kirar, S., & Soni, M. (n.d.). Glaucoma diagnosis from fundus images using convolutional neural network model: VGG-19. [Conference/Preprint]. https://s3.amazonaws.com/academia.edu.documents/Glaucoma_VGG19.pdf [5] Patel, T., Saxena, P., & Kirar, S. (2023). A literature survey on different denoising techniques in EMD–ECG using different filters. Journal of Emerging Technologies and Innovative Research, 10(8), 465–471. https://www.jetir.org/view?paper=JETIR2308515 [6] Patel, T., Saxena, P., & Kirar, S. (2023). An improved EMD-based ECG denoising method using adaptive switching mean filter. International Journal of Scientific Research in Science and Technology, 10(4), 457–467. https://ijsrst.com/IJSRST52310469 [7] Rathore, J., Saxena, P., & Kirar, S. (2022). An efficient routing protocol and neural network approach for detection and prevention of wormhole attack in wireless sensor networks. International Journal of Innovative Trends in Engineering, 74(4), 8–12. https://ijite.com/paper-details.php?paper_id=20227404 [8] Rathore, J., Saxena, P., & Kirar, S. (2022). Wormhole attack protection in WSN using modified on-demand multicast routing protocol by neural network algorithm. High Technology Letters, 28(4), 149–156.http://www.gjstx-e.cn/gallery/16-april2022.pdf [9] Wandile, P. S., Kirar, B. S., & Kirar, S. (2025). Filtering methods, assessment criteria, and prospects for image denoising. 2025 IEEE International Students\' Conference on Electrical, Electronics and Computer Science (SCEECS), 1–6. IEEE. https://doi.org/10.1109/SCEECS61528.2025.10545442 [10] Sardar, M., Niazi, M. M., Nasim, F., et al. (2024). Ensemble deep learning methods for detecting skin cancer. Bulletin of Business and Economics (BBE), 13. [11] Jicman, P. A., Smart, H., Ayello, E. A., & Sibbald, R. G. (2023). Early malignant melanoma detection, especially in persons with pigmented skin. Advances in Skin & Wound Care, 36(2), 69–77. [12] Jamil, D., Qazi, F., Shawar Agha, D., & Palaniappan, S. (2023). Classification of skin lesions using deep convolutional neural networks by applying transfer learning. Journal of Autonomous Intelligence, 6(1). [13] Almufareh, M. F., Tariq, N., Humayun, M., & Khan, F. A. (2024). Melanoma identification and classification model based on fine-tuned convolutional neural network. Digital Health, 10, 20552076241253756. https://doi.org/10.1177/20552076241253756 [14] Naqvi, M., Gilani, S. Q., Syed, T., Marques, O., & Kim, H.-C. (2023). Skin cancer detection using deep learning: A review. Diagnostics, 13(8), 1911. [15] Mahmood, T., Khan, S. H., Rashid, M., Dastgir, M., & Shah, S. A. (2023). A brief survey on breast cancer diagnosis with deep learning schemes using multi-image modalities. IEEE Access, 11, 152876–152898. [16] Abbas, M., Raza, S. H., Azam, F., & Ali, M. (2023). Comparative analysis of deep learning models for skin lesion classification. Computers in Biology and Medicine, 161, 107231. https://doi.org/10.1016/j.compbiomed.2023.107231 [17] Rahman, M. M., Afzal, M. T., Jahan, S., & Islam, M. S. (2023). Transfer learning-based deep learning approach for melanoma detection. Healthcare, 11(4), 519. [18] Tariq, M., Aslam, M., & Khan, N. (2022). Automated melanoma detection using deep learning models. Journal of Medical Imaging and Health Informatics, 12(7), 1672–1680. [19] Wang, H., Zhang, Y., Zhou, Z., Liu, L., & Hu, W. (2023). Attention-based convolutional neural network for skin cancer classification. IEEE Transactions on Medical Imaging, 42(5), 1376–1385. [20] Xiao, X., Lu, J., Zhao, Y., & Qian, C. (2022). EfficientNet-based deep learning model for melanoma recognition. Computational Intelligence and Neuroscience, 2022, 9738214. [21] Smith, J., Patel, K., & Brown, L. (2023). Role of artificial intelligence in dermatology: A systematic review. International Journal of Dermatology, 62(3), 458–472. [22] Zhang, L., Li, X., Wang, P., & Huang, J. (2023). Deep learning-based melanoma detection system using dermoscopic images. Journal of Biomedical Informatics, 138, 104185. [23] Chen, Y., Zhang, T., Liu, R., & Wang, S. (2023). Ensemble deep learning models for improving skin cancer diagnosis. Artificial Intelligence in Medicine, 142, 102514. [24] Oliveira, A. L., Silva, C. P., & Souza, D. R. (2022). Data augmentation techniques for skin lesion classification. Journal of Digital Imaging, 35(4), 829–842. [25] Kumar, R., Gupta, D., Sharma, M., & Kaur, A. (2023). A hybrid deep learning model for accurate melanoma classification. Neural Computing and Applications, 35(10), 16259–16272. [26] Li, J., He, Y., Zhang, H., & Wang, X. (2023). Real-time melanoma detection using mobile deep learning models. IEEE Transactions on Mobile Computing, 22(3), 1518–1530.

Copyright

Copyright © 2026 Tanushri Jhod, Seema Kirar. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.