Few-Shot Learning Approach To Acne Classification

Acne is a common skin condition that affects people worldwide.Correctly identifying acne types is crucial in prescribing and developing effective treatments.However, traditional classification of acne by a dermatologist is subjective, timeconsuming, and not widely accessible.To address this problem, in this paper, we propose a novel machine learning approach that achieved an impressive 91.75% accuracy in acne classification, even in a low-data scenario, using Few-shot learning (FSL) techniques.FSL is a type of machine learning designed for low-data scenarios.The FSL approach was implemented using Prototypical Networks, which utilize pre-trained CNN models to generate embeddings for the acne samples.We experimented with various CNN models in order to determine which would produce the highest accuracy.This was the GoogLeNet model.The results were compared against the performance of pre-trained CNN models and a CNN-established baseline for evaluation.The FSL approach outperformed current state-of-the-art models by 35.12% (compared to a CNN baseline) and 15.19% (compared to a pre-trained CNN model).The high accuracy achieved by this FSL approach to classifying acne shows potential of being integrated in an app or other medical systems to aid patient diagnosis by enhancing accessibility, and dermatologist diagnosis accuracy and efficiency.