Early detection of jaundice is crucial for the prevention and treatment of liver diseases; however, most people struggle to recognize symptoms at a mild stage. We aim to enhance self-monitoring capabilities by using smartphone imaging combined with machine learning for jaundice detection. Previous research by Su et al. (2021) applied deep and machine learning for jaundice prediction, yet relied on specialized color cards for color correction, resulting in higher costs and limited applicability. In response, our study proposes replacing color cards with white balance algorithms, including the white patch and gray world methods, combined with deep learning models DCCNM1 and DCCNM2, to improve accessibility and usability in jaundice detection. Evaluation results indicate that DCCNM2 achieved the best performance among models without color cards. Although its metrics slightly trail behind color card-corrected results, DCCNM2 demonstrates excellent stability and accuracy, proving its feasibility as a no-color-card alternative for jaundice screening. This method offers a convenient home-based jaundice detection solution, especially for residents in remote areas, enhancing early detection opportunities, reducing healthcare burden, and further promoting public health management.