Nail diseases, such as onychomycosis and psoriasis, are common conditions that require early and accurate diagnosis to prevent complications. This study proposes APP-Net, a novel deep learning model that integrates EfficientNetB0 with an attention pyramid pooling (APP) module, enabling precise and efficient nail disease classification using normal finger photos. The model is trained and evaluated on a curated dataset comprising normal, onychomycosis, and psoriasis categories, with augmentation applied to enhance class balance. To validate the effectiveness of APP-Net, a comprehensive comparison is conducted against state-of-the-art classification models, including VGG16, MobileNetV2, and deep hybrid learning approaches. Experimental results show that APP-Net outperforms existing models, achieving a 99.3% accuracy, the highest among all tested architectures. The APP module enhances feature representation by capturing multi-scale spatial and contextual details, improving classification performance. Furthermore, Grad-CAM visualizations illustrate APP-Net’s superior feature localization for disease-affected nail regions. Compared to traditional CNN-based methods, APP-Net demonstrates higher accuracy, robustness, and computational efficiency, making it an ideal solution for automated nail disease detection in clinical and telemedicine applications.