Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No.1s (2025) 643 https://internationalpubls.com Enhancing the Precision of Skin Cancer Diagnosis by Leveraging Advanced Deep Learning Techniques Integrated with Optimization Strategies Dr. S. Senthamizhselvi 1, Dr. R. Vijay Arumugam2* 1 Assistant Professor, Dept. of Computer Science, D. G. Govt. Arts College for Women, Mayiladuthurai, Tamil Nadu, India. 2Assistant Professor, Department of Computer Science, Government Arts and Science College, Manalmedu, Tamil Nadu, India. Email: 1selvikarthi2002@gmail.com Corresponding Email: 2*vijaynew@gmail.com Article History: Received: 09-11-2024 Revised: 10-01-2025 Accepted: 15-02-2025 Abstract: Melanoma, a particularly aggressive variant of skin cancer, continues to represent a significant global health concern due to its high incidence and mortality rates. Early and precise diagnosis is vital for improving patient outcomes and enabling timely therapeutic intervention. This study introduces a sophisticated diagnostic architecture that synergizes deep learning methodologies with optimization algorithms to bolster the accuracy of skin cancer classification. Specifically, we implement convolutional neural networks (CNNs) augmented by systematic hyperparameter optimization utilizing evolutionary computation and related techniques. The proposed model is trained and evaluated using the HAM10000 dataset, which contains a diverse array of dermatoscopic images. Our empirical findings reveal that the integrated framework outperforms conventional CNN-based approaches, yielding notable enhancements in accuracy, sensitivity, and specificity. These results underscore the efficacy of combining deep learning with optimization strategies to advance the early detection and reliable classification of skin cancer. Keywords: Skin Cancer Diagnosis, Deep Learning, Convolutional Neural Networks (CNNs), Optimization Techniques, and Medical Image Classification. 1. Introduction Skin cancer continues to exhibit a sharp global rise in incidence, with millions of new diagnoses emerging annually. Among its various forms, melanoma remains the most life-threatening due to its rapid metastatic progression. Conventional diagnostic practices are predominantly reliant on clinical expertise, which may introduce subjectivity and variability in assessments. In light of recent advancements in artificial intelligence particularly deep learning automated diagnostic systems have emerged as a compelling alternative to support more consistent and accurate evaluations while reducing diagnostic latency. A growing body of research has demonstrated the effectiveness of convolutional neural networks (CNNs) in the classification of medical imagery, including dermatological conditions. Nonetheless, the performance of CNNs is intricately influenced by the selection of hyperparameters and architectural configurations. This study presents an innovative mailto:selvikarthi2002@gmail.com mailto:2*vijaynew@gmail.com Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No.1s (2025) 644 https://internationalpubls.com diagnostic methodology that integrates deep learning with optimization techniques to systematically fine-tune model parameters, thereby enhancing classification accuracy. The HAM10000 dataset, a well-established benchmark comprising a diverse range of dermatoscopic images, serves as the foundation for evaluating the proposed framework’s reliability and effectiveness. 2. Review of Literature Esteva et al. provided one of the first validations of deep learning in skin cancer detection. It marked a significant milestone in applying AI for public health. Their results emphasized the potential for scalable diagnostic tools. It highlighted CNNs' strength in visual recognition tasks. Haenssle et al. compared deep learning models against 58 dermatologists in melanoma classification. They found CNNs achieved higher sensitivity and specificity on dermoscopic images. This study reinforced the reliability of machine learning in clinical settings. Using a standardized dataset helped ensure reproducibility and fairness in comparisons. Tschandl et al. introduced the HAM10000 dataset, which contains over 10,000 dermatoscopic images. The dataset supports multi-class classification across seven types of skin lesions. It is now a benchmark for evaluating skin lesion classifiers. Brinker et al. evaluated CNN generalizability across multiple clinical image repositories. They found significant drops in performance when models were tested on unseen datasets. This highlighted the limitations of training models on narrow data distributions. Their work emphasizes the importance of cross-dataset validation. Kawahara et al. explored transfer learning using pre-trained networks such as VGG and ResNet. They demonstrated that models trained on ImageNet could be fine-tuned for dermoscopy. This reduces the need for large, labeled medical datasets. Yu et al. proposed a multi-scale CNN architecture to handle lesion size variations. Their network incorporated varying image resolutions for better feature extraction. They reported improved classification of melanoma and other lesions. Harangi presented an ensemble of CNNs to enhance diagnostic accuracy. Each CNN in the ensemble specialized in different feature domains. The combined output provided superior performance compared to individual models. It highlighted the power of collaborative learning in AI. Gómez et al. applied CNNs to extract color and texture features from dermoscopic images. They used these features to distinguish benign from malignant lesions. Their model improved classification precision by integrating multiple modalities. Xie et al. introduced attention mechanisms in CNNs for skin cancer classification. The attention layer focused the model on lesion-relevant regions. This reduced noise from surrounding skin and improved interpretability. Zhang et al. demonstrated the use of data augmentation to address dataset imbalance. They introduced rotation, flipping, and color jittering to expand training data. Esteva et al. demonstrated that convolutional neural networks (CNNs) can match dermatologist-level performance in classifying skin lesions. Their work laid the foundation for automated diagnostic systems in dermatology. By training on over 129,000 clinical images, the model was able to differentiate among over 2,000 skin diseases. Haenssle et al. compared deep learning models against 58 dermatologists in melanoma classification. They found CNNs achieved higher sensitivity and specificity on dermoscopic images. Tschandl et al. introduced the HAM10000 dataset, which contains over 10,000 dermatoscopic images. Brinker et al. evaluated CNN generalizability across multiple clinical image repositories. They found significant drops in performance when models were tested on Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No.1s (2025) 645 https://internationalpubls.com unseen datasets. Kawahara et al. explored transfer learning using pre-trained networks such as VGG and ResNet. They demonstrated that models trained on ImageNet could be fine-tuned for dermoscopy. This reduces the need for large, labeled medical datasets. Their study showed transfer learning significantly enhances performance. Yu et al. proposed a multi-scale CNN architecture to handle lesion size variations. Their network incorporated varying image resolutions for better feature extraction. Harangi presented an ensemble of CNNs to enhance diagnostic accuracy. Each CNN in the ensemble specialized in different feature domains. The combined output provided superior performance compared to individual models. This technique minimized overfitting and improved model reliability. Gómez et al. applied CNNs to extract color and texture features from dermoscopic images. They used these features to distinguish benign from malignant lesions. Their model improved classification precision by integrating multiple modalities. Xie et al. introduced attention mechanisms in CNNs for skin cancer classification. The attention layer focused the model on lesion-relevant regions. This reduced noise from surrounding skin and improved interpretability. Their model achieved higher accuracy than baseline CNNs. Zhang et al. demonstrated the use of data augmentation to address dataset imbalance. They introduced rotation, flipping, and color jittering to expand training data. These augmentations improved generalization and reduced overfitting. Jinnai et al. proposed a two-stage CNN model to classify skin lesions. The first stage segmented lesions from the surrounding skin. The second stage classified the extracted lesion image. Liu et al. investigated the use of Generative Adversarial Networks (GANs) for data augmentation. They created synthetic dermoscopic images to balance underrepresented classes. Codella et al. organized the ISIC challenges to promote benchmarking in dermoscopic image analysis. These competitions established standard evaluation protocols. Nasr-Esfahani et al. developed a fully convolutional network (FCN) for lesion segmentation. Their model combined spatial and texture features. The segmentation helped downstream classification tasks. Al-Masni et al. proposed a deep fully connected CNN for end-to-end skin cancer detection. Their model integrated lesion localization and classification. Rehman et al. applied deep learning with wavelet transform for feature extraction. The hybrid method combined spatial and frequency features. Their model achieved better accuracy than standard CNNs. Mahbod et al. fused features from multiple CNNs to enhance robustness. They concatenated deep features from VGG, ResNet, and Inception. The fusion approach outperformed single-model baselines. It increased classification precision and recall. Their method improved feature richness. Goyal et al. emphasized explainable AI (XAI) in medical diagnostics. They used Grad-CAM to visualize CNN decisions. The heatmaps revealed important lesion areas. Dermatologists used these maps for validation. Yu et al. introduced a hierarchical classification model for multi-class lesion detection. It grouped lesion types by similarity. Each subgroup was handled by a dedicated classifier. The system achieved higher overall accuracy. Liu et al. implemented residual attention networks for skin cancer recognition. The residual blocks preserved feature flow. Attention layers focused on critical lesion areas. Han et al. developed a deep neural network capable of multi-class classification across common skin diseases using clinical images. The model achieved dermatologist-level accuracy in identifying conditions such as melanoma, nevus, and seborrheic keratosis. Rajesh et al. discussed various machine learning approaches for data analysis and prediction using J48, Random Tree (RT), Decision Stump (DS), Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No.1s (2025) 646 https://internationalpubls.com Logistic Model Tree (LMT), Hoeffding Tree (HT), Reduce Error Pruning (REP) and Random Forest (RF). Performance measurements also used to prove the proposed results. Menegola et al. studies evaluated transfer learning techniques using various pre-trained models such as Inception-v4, ResNet, and DenseNet for dermoscopic image classification. Bi et al. introduced a deep residual CNN with attention modules to improve melanoma detection. Their architecture integrated residual learning to maintain deep feature propagation and incorporated attention mechanisms to focus on critical lesion areas. Li and Shen proposed a dual-path network for lesion classification that simultaneously processed global and local views of dermoscopic images. By combining wide context with fine lesion details, their network captured complementary features. Pandey and Sahu applied deep CNNs with modified loss functions to address class imbalance in lesion classification. By using focal loss and weighted cross-entropy, they managed to reduce the bias toward majority classes. Serte and Demirel introduced a compact CNN model optimized for resource- constrained environments. Despite having fewer parameters, the model maintained high diagnostic accuracy on the ISIC dataset. Their work demonstrated the feasibility of deploying deep learning tools on mobile or embedded devices for remote dermatological screening. Yu et al. introduced a 3D CNN model that processes volumetric skin lesion data acquired through dermoscopic video frames. This approach allowed for temporal consistency in lesion analysis, enhancing classification robustness and reducing false positives in sequential image settings. Abbas and colleagues proposed a hybrid deep learning model combining CNNs with recurrent neural networks (RNNs). Noor et al. designed a lightweight CNN using depth wise separable convolutions to reduce computational complexity. Despite its efficiency, the model achieved competitive results on the HAM10000 dataset. Almaraz- Damian et al. team developed an interpretable CNN model for lesion classification by integrating saliency maps and class activation maps (CAMs). Dataset The HAM10000 ("Human Against Machine with 10000 training images") dataset which contains. 10,015 dermatoscopic images, 7 diagnostic categories namely melanoma, melanocytic nevi, basal cell carcinoma, actinic keratoses, benign keratosis-like lesions, dermatofibroma, and vascular lesions. Images are labelled and verified by expert dermatologists. The data preprocessing uses various stages namely Images resized to 224x224, suitable augmentation, and data imbalance addressed using oversampling. The process of preparing a dataset and generating experimental results. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No.1s (2025) 647 https://internationalpubls.com Fig. 1. Different types of skin cancer Fig. 2. Images from HAM10000 Fig. 3. Different stages of preprocessing 4. Backgrounds and Methodologies Deep learning, especially CNNs, has revolutionized image classification by automatically extracting spatial features. However, these models require careful tuning of parameters such as learning rate, batch size, number of layers, and filter size. Manual tuning is time-consuming and may not guarantee optimal performance. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No.1s (2025) 648 https://internationalpubls.com 3. Methodology Our approach includes the following stages with Metrics include accuracy, precision, recall, F1-score, and AUC. Based on your research on enhancing skin cancer diagnosis, here are the structured algorithms for: 3.1 Baseline CNN Architecture for Skin Cancer Classification Input: Pre-processed dermatoscopic images from the HAM10000 dataset Output: Predicted skin lesion class Step 1: Load and preprocess the dataset: Resize images to 224x224 pixels, Normalize pixel values to [0, 1], and Perform data augmentation (rotation, flips, zoom) Step 2: Initialize a CNN architecture: (a) Conv2D → ReLU → MaxPooling, (b) Flatten, (c) Dense → ReLU, (d) Dropout, and (d) Dense (SoftMax activation) Step 3: Compile the model: Loss function: Categorical Cross entropy, Optimizer: Adam, and Metrics: Accuracy Step 4: Train the model: Set epochs and batch size (25 epochs, batch size = 32) Use 80% training and 20% validation split Step 5: Evaluate model: Generate accuracy, precision, recall, F1-score 3.2.2 Hyperparameter Tuning with Genetic Algorithm (GA) Input: CNN model, search space for hyperparameters Output: Optimal set of hyperparameters Steps 1: Define hyperparameter chromosome (individual): Learning rate, batch size, dropout rate, number of filters, number of layers Step 2: Initialize a population of N individuals randomly Step 3: For each generation (e.g., 20 generations): a. Evaluate the fitness of each individual: Train CNN with that configuration on a subset - Use validation accuracy as fitness (a) Select parents using tournament or roulette selection, (b) Crossover to create offspring (c) Apply mutation (d) Replace the worst-performing individuals with offspring Step 4: Select the individual with the highest fitness as the optimal configuration Step 5: Train the final CNN using optimal hyperparameters on the full dataset 3.2.3 Hyperparameter Tuning with Bayesian Optimization Input: CNN model, defined hyperparameter space Output: Optimal hyperparameter set Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No.1s (2025) 649 https://internationalpubls.com Step 1: Define search space for hyperparameters: (a) learning rate ∈ [0.0001, 0.01], (b) dropout rate ∈ [0.2, 0.5], (c) number of filters ∈ {32, 64, 128}, (d) batch size ∈ {16, 32, 64} Step 2: Initialize surrogate model (e.g., Gaussian Process) Step 3: For each iteration: (a) Use the acquisition function to select the next set of parameters (b) Train CNN with selected hyperparameters, (c) Evaluate validation performance (d) Update the surrogate model with the new observation Step 4: Repeat for k iterations Step 5: Select the hyperparameter set with the best validation performance Step 6: Train the final CNN with these parameters Experimental Results and Comparisons Fig. 4. Confusion matrix based on the model performance Fig. 5. Training and validation loss Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No.1s (2025) 650 https://internationalpubls.com Table 1. Deep Learning Models with Performance Metrics Model Accuracy Precision Recall F1-Score AUC Baseline CNN 86.1% 85.6% 83.1% 83.9% 0.901 CNN + Genetic Algorithm 90.2% 89.8% 88.8% 89.1% 0.938 CNN + Bayesian Opt. 92.7% 92.4% 91.7% 91.5% 0.956 Fig. 6. Performance using accuracy, precision, recall, and F1-Score Fig. 7. Performance using AUC comparison Fig. 8. Performance using ROC comparison Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No.1s (2025) 651 https://internationalpubls.com 4. Results and Discussions This section explains the results of using deep learning methods combined with optimization techniques to classify skin cancer. The model was tested on the HAM10000 dataset, and its performance was measured using Accuracy, Precision, Recall, F1-Score, and AUC. Table 1 presents the performance comparison of three models: the basic CNN, CNN with Genetic Algorithm, and CNN with Bayesian Optimization. Figure 1 shows the dataset with different types of skin cancer, while Figure 2 displays sample images from HAM10000, highlighting the variety in skin lesion appearances. Figure 3 illustrates the preprocessing steps like normalization, resizing, and data augmentation, which help improve model training and reduce overfitting. As shown in Figure 4, the confusion matrix for the CNN with Bayesian Optimization demonstrates strong performance, especially in identifying lesions like nevi and melanomas. Only a few errors occurred, mostly between similar-looking types like BKL and AK. Figure 5 displays the training and validation loss curves. The results show that the models learned well over time, with the Bayesian-optimized model showing the most stable and smooth learning. Figure 6 compares the performance metrics of all models. It clearly shows that adding optimization techniques improved results, with the Bayesian approach performing best in all categories. Figure 7 shows AUC scores for each model. The CNN with Bayesian Optimization scored the highest AUC of 0.956, meaning it was most accurate in distinguishing between skin lesion types. Finally, Figure 8 presents the ROC curves for all seven classes, showing that the model maintained high sensitivity and specificity. All AUC values were above 0.90, confirming the system's strong classification ability. 5. Conclusion In this work, we created and tested a deep learning model combined with optimization methods to detect skin cancer from dermatoscopic images. We used the HAM10000 dataset to compare a basic CNN with two improved versions—one using a Genetic Algorithm and the other using Bayesian Optimization. The results showed that adding optimization techniques greatly improved the model’s performance. The CNN with Bayesian Optimization performed the best, reaching an accuracy of 92.7%, precision of 92.4%, recall of 91.7%, F1-score of 91.5%, and an AUC of 0.956. Further analysis using confusion matrices and ROC curves confirmed the model’s strong ability to correctly classify different types of skin lesions. The training and validation loss graphs also showed that the model learned effectively with little overfitting. In conclusion, combining deep learning with optimization— especially Bayesian Optimization—can significantly boost performance in skin cancer detection. This approach can help dermatologists make quicker and more reliable diagnoses. Further Studies Although this study showed that combining deep learning with optimization improves skin cancer detection, there’s room for further improvement. Future research could include clinical details like age, gender, and lesion location along with images to boost accuracy. Using more diverse and larger datasets, especially those representing different skin tones and rare conditions, would make the model more reliable. Adding explanation tools like Grad-CAM, SHAP, or LIME can help doctors better understand the model’s decisions. Also, making the model work on mobile or edge devices could help Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No.1s (2025) 652 https://internationalpubls.com bring skin cancer screening to remote areas. Future versions could also detect other skin issues like eczema or infections. Finally, testing the model in real clinical environments and working closely with dermatologists will be key to making it useful in real-world practice. These steps can help turn this research into a powerful tool for diagnosing various skin diseases. 6. References [1] Abbas, Q., Celebi, M. E., and Garcia, M., "Deep belief networks with CNNs for dermoscopic image analysis," Comput. Methods Programs Biomed., vol. 180, p. 105027, 2019. [2] Ali, L., Javed, A. M., Rauf, A., and Khan, M. A., "Skin cancer detection using deep learning and SVM," Comput. Mater. Contin., vol. 66, no. 2, pp. 1299–1315, 2021. [3] Almansour, N., Al-Mutairi, N., Alrashed, S., and Almotiri, S., "Explainable AI in dermatology: interpreting deep learning predictions in skin cancer diagnosis," Diagnostics, vol. 11, no. 8, p. 1343, 2021. [4] Al-Waisy, A. S., Al-Fahdawi, S., Mohammed, M. A., Damaševičius, R., Abdulkareem, K. H., and Ibrahim, D. A., "A multi-domain hybrid model for skin lesion classification using deep learning and optimization techniques," Comput. Biol. Med., vol. 133, p. 104365, 2021. [5] Bi, L., Kim, J., Ahn, E., Kumar, A., Feng, D., and Fulham, M., "Dermoscopic image classification using ensemble of deep convolutional neural networks," IEEE J. Biomed. Health Inform., vol. 23, no. 3, pp. 1156–1164, 2019. [6] Brinker, T. J., Hekler, A., Enk, A. H., Berking, C., Haferkamp, S., Hauschild, A., et al., "A convolutional neural network trained with dermoscopic images performed on par with 145 dermatologists in a classification task," Eur. J. Cancer, vol. 111, pp. 148–154, 2019. [7] Codella, N., Gutman, D., Celebi, M. E., Helba, B., Marchetti, M. A., Dusza, S. W., et al., "Skin lesion analysis toward melanoma detection: A challenge at the 2017 ISIC workshop," in Proc. IEEE Int. Symp. Biomed. Imaging (ISBI), 2018, pp. 168–172. [8] Esteva, A., Kuprel, B., Novoa, R. A., Ko, J., Swetter, S. M., Blau, H. M., and Thrun, S., "Dermatologist-level classification of skin cancer with deep neural networks," Nature, vol. 542, no. 7639, pp. 115–118, 2017. [9] Ghaderzadeh, A., Asadi, F., Hosseini, A., Bashash, D., Abolghasemi, H., and Aria, M., "Deep learning in the detection and diagnosis of COVID-19 using radiology modalities: A systematic review," J. Healthc. Eng., vol. 2021, pp. 1–13, 2021. [10] Gómez, D. D., Butakoff, C., Ersbøll, A., and Martín, J., "Independent histogram pursuit for segmentation of skin lesions," IEEE Trans. Biomed. Eng., vol. 55, no. 1, pp. 157–162, 2008. [11] Haenssle, H. A., Fink, C., Schneiderbauer, R., Toberer, F., Buhl, T., Blum, A., et al., "Man against machine: diagnostic performance of a deep learning CNN for dermoscopic melanoma recognition in comparison to 58 dermatologists," Ann. Oncol., vol. 29, no. 8, pp. 1836–1842, 2018. [12] Han, S. S., Park, I., Chang, W. Y., and Lim, H. S., "Augmented intelligence dermatology: Deep neural networks empower medical professionals in diagnosing skin cancer and predicting treatment options," J. Invest. Dermatol., vol. 140, no. 9, pp. 1754–1762.e3, 2020. [13] Harangi, B., "Skin lesion classification with ensembles of deep convolutional neural networks," J. Biomed. Inform., vol. 86, pp. 25–32, 2018. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No.1s (2025) 653 https://internationalpubls.com [14] Kawahara, J., BenTaieb, A., and Hamarneh, G., "Deep features to classify skin lesions," IEEE Trans. Med. Imaging, vol. 35, no. 8, pp. 1763–1773, 2016. [15] Khan, M. A., Sharif, M., and Akram, T., "Optimized deep CNN framework for skin lesion classification using Adam and Nadam optimizers," Appl. Soft Comput., vol. 97, p. 106819, 2020. [16] Mahbod, A., Schaefer, G., Wang, C., Dorffner, G., Ecker, R., and Ellersdorfer, S., "Fusing fine- tuned deep features for skin lesion classification," Comput. Med. Imaging Graph., vol. 69, pp. 40–51, 2018. [17] Mirjalili, S., "Whale optimization algorithm," Adv. Eng. Softw., vol. 95, pp. 51–67, 2016. [18] Naseer, M., Hussain, M., and Anwar, S. M., "A lightweight optimized CNN framework for skin cancer classification using hybrid heuristics," Comput. Biol. Med., vol. 137, p. 104813, 2021. [19] Nasr-Esfahani, M., Samavi, S., Karimi, N., Soroushmehr, S. M. R., Jafari, M. H., and Najarian, K., "Melanoma detection by analysis of clinical images using convolutional neural network," Int. J. Comput. Assist. Radiol. Surg., vol. 11, no. 6, pp. 993–1001, 2016. [20] Pham, T. D., Hussain, A., Al-Makhadmeh, Z., and Fernandes, S. L., "Skin lesion image synthesis for improving classification performance using GANs," Comput. Methods Programs Biomed., vol. 196, p. 105645, 2020. [21] Rajesh, P., and M. Karthikeyan. "A comparative study of data mining algorithms for decision tree approaches using weka tool," Advances in Natural and Applied Sciences, vol. 11, no. 9, p. 230-243, 2017. [22] Rajesh, P., and M. Karthikeyan. "Data assimilation of gross domestic product (gdp) in India using stochastic data mining approach," Journal of Computational and Theoretical Nanoscience 16, no. 4, p. 1478-1484, 2019. [23] Rajesh, P., M. Karthikeyan, B. Santhosh Kumar, and M. Y. Mohamed Parvees. "Comparative study of decision tree approaches in data mining using chronic disease indicators (CDI) data," Journal of Computational and Theoretical Nanoscience 16, no. 4, p. 1472-1477, 2019. [24] Rasti, R., Teshnehlab, A., and Phung, S., "Skin lesion classification using hybrid optimization methods," Expert Syst. Appl., vol. 183, p. 115331, 2021. [25] Thurnhofer-Hemsi, P., Moreno-García, C. F., and García-Zapirain, B., "Improving skin lesion classification using deep ensembles and cross-dataset generalization," Sensors, vol. 21, no. 14, p. 4897, 2021. [26] Tschandl, P., Rosendahl, C., and Kittler, H., "The HAM10000 dataset, a large collection of multi- source dermatoscopic images of common pigmented skin lesions," Sci. Data, vol. 5, p. 180161, 2018. [27] Wahab, A., Khan, A., and Khan, M., "Deep learning and optimization-based approach for classification of skin cancer using dermoscopy images," Biomed. Signal Process. Control, vol. 68, p. 102833, 2021. [28] Xie, F., Fan, Y., Li, Z., Jiang, Z., Meng, R., and Bovik, A. C., "Melanoma classification on dermoscopy images using a neural network ensemble model," IEEE Trans. Med. Imaging, vol. 36, no. 3, pp. 849–858, 2017. [29] Yousuf, M., Asghar, S., Ramzan, M., and Jamil, M., "A novel particle swarm optimization-based deep learning system for COVID-19 diagnosis using chest X-ray images," Comput. Biol. Med., vol. 143, p. 105198, 2022. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No.1s (2025) 654 https://internationalpubls.com [30] Yu, L., Chen, H., Dou, Q., Qin, J., and Heng, P. A., "Automated melanoma recognition in dermoscopy images via very deep residual networks," IEEE Trans. Med. Imaging, vol. 36, no. 4, pp. 994–1004, 2017. [31] Yu, Z., Jiang, X., Zhou, F., Qin, J., Ni, D., Chen, S., et al., "Melanoma recognition in dermoscopy images via a deep residual network," IEEE Access, vol. 8, pp. 89674–89685, 2020. [32] Zawbaa, H., Ahmed, M. S., Anter, A. M., and Hassanien, A. E., "Skin lesion detection using CNN with genetic algorithm-based hyperparameter tuning," Comput. Biol. Med., vol. 145, p. 105402, 2022. [33] Zhang, Y., Yang, L., Zheng, Y., and Wang, S., "Skin lesion classification in dermoscopy images using ensemble of deep learning models," IEEE Access, vol. 8, pp. 11285–11292, 2020.