In many low-resource settings, the debilitating effects of arsenicosis remain a critical public health concern. This condition leads to serious health problems, including skin lesions, skin cancer, cardiovascular disease, and neurological impairment. Visual inspection is the standard practice for diagnosing arsenicosis; however, it is often subjective and can lead to variability in diagnosis. To address the urgent need for efficient arsenicosis skin detection, we propose ArsenicSkinNet, adapted from the EfficientNet architecture for the automatic classification of arsenicosis skin disease. ArsenicSkinNet is a state-of-the-art technique based on the EfficientNet neural network architecture and transfer learning, known for achieving high performance with fewer computational resources compared to traditional models. Our results show that ArsenicSkinNet achieved a classification accuracy of 99.61% on the arsenicosis skin lesion dataset. Our study demonstrates that deep learning could significantly improve the capacity for early detection and management of arsenicosis, thereby mitigating its impact on affected populations in low-resource settings.