Abstract
This study presents the development of a low-cost, portable device for non-invasive skin monitoring that integrates both impedance sensing and imaging modalities to assess skin moisture levels and classify skin types. To enable real-time analysis on resource-constrained platforms, lightweight machine learning algorithms were employed for both regression and classification tasks. For the moisture prediction task, experimental results demonstrate that the Random Forest (RF) algorithm outperformed Linear Regression and Multilayer Perceptron (MLP), achieving the highest accuracy, with impedance-based data yielding better performance than image-based inputs. In the skin type classification task, the MLP model trained on handcrafted features outperformed a convolutional neural network (CNN) applied to raw images, highlighting the effectiveness of feature-engineered approaches. The proposed system shows strong potential for applications in personalized skincare, dermatological assessment, and portable health monitoring.