A Mobile Application for Detecting Skin Cancer Using Convolutional Neural Networks and Random Forest Algorithm

Abstract

As Skin is the largest part of body, skin Cancer is the most hazardous disease in the human body. Skin cancer occurs, when malignant tumors form on the skin and it gives protection to the body from heat, Ultraviolet rays and injury for internal organs. The Skin has many layers; the two main layers are epidermis and dermis. The epidermis is the place where malignant cells will beappearing. Melanocytes make melanin and appear in epidermis. Skin cancer diagnosis includes: full skin examination, a biopsy which includes shave biopsy, punch biopsy and excision biopsy. These tests take longer time, while the malignant cells will spread to nearby cells. Advance diagnosis of melanoma is very vital in saving a patient's lives. Although melanoma is a Hazardous disease of skin cancer, it can be treated only if it diagnosed in advance. To address the need for a diagnostic tool that is rapid, accurate, and cost-effective, this study explores the use of automated computer-based methods for skin cancer assessment. In the proposed approach, image pre-processing is applied to the input data to reduce noise and eliminate unwanted visual artifacts. A Convolutional Neural Network (CNN) model is then used to enhance the system’s ability to classify and detect skin lesions.Region-based CNN (R-CNN) techniques have demonstrated strong performance in handling large image datasets, achieving reported accuracies in the mid-80% range in prior research. Their precision and consistency make them suitable for assisting with skin cancer identification and lesion categorization. Random Forest (RF) is a widely used ensemble learning algorithm designed for classification and regression tasks. It was introduced as an extension of decision tree algorithms to improve prediction accuracy, reduce over-fitting, and increase robustness. Because of its ability to handle high-dimensional data and noisy features, the algorithm has been extensively applied in medical diagnosis, bioinformatics, image analysis, and other domains requiring reliable predictive performance. Building on these methods, our solution incorporates a cloud-supported mobile application that captures images of suspicious lesions and provides an indication of whether the tumour is likely benign or malignant.