Skin diseases represent a worldwide concern in terms of public health, and early detection is critical for effective therapy. Automated dermatological analysis is a potential answer with deep learning, with increased accuracy and efficiency. In this work, a new model is proposed with a feature extraction using EfficientNetB0 and a custom Convolutional Neural Network (CNN) for high-classification performance. Squeeze-and-Excitation (SE) blocks and techniques such as random flipping, rotation, zoom, contrast, and translation for enhancing generalization capabilities are utilized in the model. EfficientNetB0 is partially fine-tuned with all but its last 50 layers frozen for a proper feature transfer and task-specific training balancing. It utilizes scheduled learning rate decay, early stopping, and model checkpointing with an Adam optimizer for efficient training and minimizing overfitting. For testing, a disease skin dataset is utilized, and performance evaluation is conducted in terms of accuracy, precision, recall, and F1-score, and analysis through a confusion matrix is performed for evaluation of proposed model performance. Robust classification accuracy in experimental analysis proves efficiency in proposed hybrid model for disease skin detection, and its real-life clinical application is a positive sign with its potential to aid in quick and accurate diagnosis for dermatologists.