Skin Cancer Detection using Ensemble Learning
J. Avanija, C. Chandra Mohan Reddy, C. Sri Chandan Reddy, D. Harshavardhan Reddy, Thumma Narasimhulu, Napa Venkata Hardhik
Abstract
Skin cancer is a serious and life-threatening condition resulting from the unrestrained multiplication of abnormal cells present in the extreme layer of the skin. This happens when DNA damage goes unrepaired and mutations arise causing these cells to increase rapidly, thus developing malignant tumors. Skin cancer can be classified into four categories: melanoma, basal cell carcinoma (BCC), squamous cell carcinoma (SCC) and Merkel cell carcinoma (MCC). In the past, machine learning has been used to detect this type of cancer by analyzing protein sequences and various types of imaging. Machine Learning usually requires manually-engineered features, which can be laborious and time-consuming. Luckily, Deep Learning provides a solution to this issue by allowing an automatic feature extraction process. In this research, convolutional-based deep neural networks were used to identify cancer affected skin making use of the ISIC Skin Cancer Dataset from Kaggle. Timely and accurate detection of cancer is crucial and any errors made can be quite severe, so this study has provided a reliable solution. The individual machine learning models for cancer detection are not as reliable and accurate as required. Furthermore, the challenge of generalizability of the ML algorithms developed in recent times to address real-world scenarios is also a concern. The performance of algorithms may be affected by the variations in lighting, image quality, and other factors that can impact image analysis. So, from data selection and augmentation to regularization and model architecture, this study has ensured that the model can perform well on new and unseen data. To improve the accuracy of predictions for sensitive issues like cancer detection, combining the decisions of individual learners can be beneficial. Therefore, the research suggests utilizing ensemble of deep learners to produce optimal output. From this, the factors important to determine the classification of skin cancer are identified.