A Recommendation System for Skincare Products Based on Acne Severity Detection Using Image Processing Techniques

Rajendra Janapati Mulki, Dr. techn. Aufaclav Zatu Kusuma Frisky, S.Si., M.Sc.

2025 | Skripsi | ELEKTRONIKA DAN INSTRUMENTASI

    Industri perawatan kulit global telah mengalami pertumbuhan yang signifikan dalam beberapa dekade terakhir. Peningkatan jumlah produk yang tersedia di pasaran, disertai dengan bertambahnya konsumen yang mencari solusi efektif untuk berbagai permasalahan kulit, mendorong kebutuhan akan pendekatan yang lebih tepat sasaran. Banyak individu, khususnya yang memiliki kondisi kulit sensitif atau bermasalah, masih mengandalkan metode trial-and-error yang bersifat tidak efisien dan berbiaya tinggi. Di samping itu, keterbatasan akses terhadap layanan dermatologi di wilayah-wilayah terpencil turut memperbesar tantangan dalam pengambilan keputusan perawatan kulit yang tepat.

    Penelitian ini bertujuan untuk merancang dan mengembangkan sebuah aplikasi mobile yang mampu memberikan rekomendasi perawatan kulit secara personal dengan memanfaatkan teknologi Artificial Intelligence (AI). Pendekatan yang digunakan melibatkan integrasi teknik Computer Vision dan Deep Learning, khususnya Convolutional Neural Networks (CNN) dan Transfer Learning. CNN digunakan sebagai model dasar dalam tugas klasifikasi citra, seperti deteksi jenis kulit dan klasifikasi tingkat keparahan jerawat. Transfer Learning diimplementasikan menggunakan arsitektur EfficientNet-B0 untuk mengekstraksi fitur-fitur penting dari citra kulit, mencakup jerawat, jenis kulit, dan warna kulit. Selain itu, algoritma K-Means Clustering digunakan pada tahap praproses citra guna meningkatkan akurasi dalam klasifikasi warna kulit. Sistem rekomendasi dibangun dengan pendekatan content-based filtering berdasarkan data produk perawatan kulit dan kosmetik.

    Hasil eksperimen awal menunjukkan bahwa pendekatan berbasis AI memiliki potensi yang tinggi dalam meningkatkan akurasi analisis dermatologis dan efisiensi pemberian rekomendasi produk. Dengan memanfaatkan sumber daya secara optimal, aplikasi ini diharapkan mampu menjadi solusi yang lebih efisien, terjangkau, dan mudah diakses bagi pengguna, serta mengurangi ketergantungan terhadap metode coba-coba dalam perawatan kulit.

    The global skincare industry has had a rapid expansion throughout the years. An overwhelming amount and variety of products, and a growing number of new consumers seeking effective solutions for their various personal skin conditions. Many individuals, most of them having sensitive or problematic skin, rely on trial-and-error methods, which can be costly and ineffective. Underserved areas often get limited access to dermatological care as well, exacerbating the challenge of making informed skincare decisions. 

    Addressing all these issues as one, the need for a mobile application that can provide personalized skincare recommendation is needed. Integrating Computer Vision (CV) techniques along with deep learning models such as Convolutional Neural Networks (CNN) and Transfer Learning, this research aims to kill two birds with one stone by offering accurate skin analysis while simultaneously recommending suitable skincare products. Convolutional Neural Networks is used as the backbone for image classification tasks, including skin type detection and acne severity classification, Transfer Learning is used with Efficient Net B0, a pre-trained deep learning model to extract meaningful skin-related features (acne, skin type, skin tone), and K-Means Clustering is used during image preprocessing to help in enhancing skin tone classification. A recommendation system is also curated using content-based filtering from various skincare and makeup datasets.

    Previous experimental results have indicated that AI-driven skincare solutions can achieve high accuracy in acne detection and skin classification. The primary goal of this research is to create an application that improves the accuracy of dermatological assessments using AI-powered recommendations while on limited resources and assessing other models via comparison. This reduces reliance on trial-and-error methods, making personalized skincare more efficient, accessible and user-friendly.

Kata Kunci : EfficientNet B0, K-Means Clustering, Recommendation System, Image Preprocessing