Melanoma, the lethal skin cancer, is triggered by melanocytes or cells producing pigment. According to Cancer Research UK, 2341 people died due to melanoma between 2017 and 2019 in the UK. It can, however, be cured, and 87% of cases survive. Melanoma is mostly diagnosed by biopsy and histopathological examination. The deep learning community has become interested in recent non-invasive cancer screening techniques that involve imaging pigmented skin lesion images. Convolution and attention modules are used by the current convolutional neural network (CNN) and transformer models to identify features and make judgments. A few limitations of these models are their complexity, lengthy training times, and high memory and graphics processing unit utilization. We have used a mixer architecture technique based on multilayer perceptron (MLP) to get around these constraints. To do this, an architecture called an MLP mixer uses patches for the input images, which are subsequently supplied to a mixer layer after being transformed into a shape resembling a table. Two MLP blocks make up the mixer layer. The first, named token mixing, aggregates channels where a feature appears in the image across patches to detect it. The second block associates the patch with the channel by searching for features in it. It also employs skip connections, layer norm, and Gaussian error linear unit. The proposed model used observational settings of parameters such as dropouts, patch size, MLP blocks, and layers to train the model for skin lesion images to get a homogeneous input. The experiment’s top model attains 87.33% accuracy, 87.3% F1-score, 87.2% recall, and 87.6% precision for binary classification on the “Human Against Machine with 10000 training images” dataset which were better contrary to existing CNN and transformer models. The proposed model is lightweight requiring less computational resources which makes it better for real-time applications.