Effect of Image Downsizing and Color Reduction on Skin Cancer Pre-screening

Every year, the skin cancer burden is increasing due to ultraviolet exposure caused by the gradual thinning of Earth's ozone layer. One of the main methods to detect skin lesion is using image processing techniques, including the use of machine learning. This research tries to explore the effect of skin cancer image downsizing and color dimensional reduction using $k$ -means clustering on skin lesion pre-screening. The contribution of this research is that for skin cancer pre-screening using CNN, the optimal combination that can be used in terms of accuracy (training and validation); image size; the number of colors; and computing time, is the combination of 8 colors and $200\times 150$ pixels image. The significance of this study is that the penetration of smartphones as low-cost computation resources are tremendous, particularly in Indonesia. This device can be used to detect the skin lesion in an easy way due to it is already equipped with a camera, processor, memory, and other computing peripherals.