Browsing by Author "MAHIR SHAHRIAR ABIR"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    Open Access
    A Lightweight Pruned DCNN Model with XAI for Skin Cancer Classification
    (North South University, 2024-08) MEHEDI HASAN; MAHIR SHAHRIAR ABIR; ABU TAYEB MOHAMMAD AYON; Dr. Sifat Momen; 2011425042; 2013640042; 2012547042
    Skin cancer is one of the most destructive types of cancer due to its immediate arrival and the potential for rapid spread. Sometimes, traditional diagnostic methods don’t react immediately because they need dermatologist expertise, some automation tools and time. This study explores one of the areas of artificial intelligencedeep learning, specifically Convolutional Neural Networks (CNNs), to enhance the efficiency and accuracy of skin cancer diagnosis. In this article, we introduce a lightweight pruned deep Convolutional Neural Network (dCNN) based method for skin cancer detection. With significantly lower trainable parameters, it surpasses recent state-of-the-art methods and shows competitive performance against 11 CNN-based pre-trained models, including Densenet121, DenseNet201, EfficientNetB2, InceptionV3, MobileNetV2, ResNet50, ResNet50V2, ResNet101V2, VGG16, VGG19, Xception. The proposed model demonstrates an accuracy of 98.07% with a notable precision of 98.15%, recall of 98.07%, and an F2 score of 98.02%. Even model interpretability perfectly works with Grad-Cam++. The proposed pruned model not only gets higher accuracy but also reduces computational cost, making it suitable for deploy on any device.