Prediksi Kecocokan Produk Skincare terhadap Jenis Kulit Menggunakan Knowledge Graph dan Relational Graph Attention Network

Mustofa, Juninda (2026) Prediksi Kecocokan Produk Skincare terhadap Jenis Kulit Menggunakan Knowledge Graph dan Relational Graph Attention Network. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5002221037-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Pemilihan produk skincare yang sesuai dengan jenis kulit merupakan tantangan yang kompleks karena adanya variasi kandungan bahan aktif serta perbedaan karakteristik biologis setiap jenis kulit. Ketidaksesuaian penggunaan produk skincare dapat menyebabkan iritasi, jerawat, maupun gangguan pada fungsi lapisan pelindung kulit. Selain itu, hubungan antara produk, kandungan bahan, kategori produk, dan jenis kulit bersifat kompleks sehingga diperlukan pendekatan yang mampu merepresentasikan hubungan antarentitas secara terstruktur untuk menghasilkan prediksi kecocokan yang lebih akurat dan mudah diinterpretasikan. Oleh karena itu, penelitian ini menggunakan Knowledge Graph untuk memodelkan keterkaitan semantik antarentitas berupa produk skincare, kategori produk, kandungan bahan, dan jenis kulit dalam bentuk graf terarah, serta Relational Graph Attention Network (RGAT) untuk mempelajari representasi graf dengan mempertimbangkan jenis relasi melalui mekanisme attention. Input model berupa representasi graf yang terdiri atas node embedding, edge index, dan edge type, sedangkan output model berupa prediksi status kecocokan (suitable atau not suitable) antara produk skincare dan jenis kulit tertentu. Hasil prediksi menunjukkan bahwa model mampu mengidentifikasi kecocokan produk SKINTIFIC 5X Ceramide Barrier Moisture Gel sebagai suitable untuk kulit berminyak berdasarkan hubungan antara kandungan bahan, kategori produk, dan jenis kulit yang direpresentasikan dalam Knowledge Graph. Hasil evaluasi menunjukkan bahwa model memperoleh nilai accuracy sebesar 84,28%, precision sebesar 85,05%, recall sebesar 83,18%, F1-score sebesar 84,10%, dan ROC AUC sebesar 91,64%. Selain menghasilkan prediksi kecocokan, model juga memberikan nilai attention weight yang menunjukkan relasi paling berpengaruh terhadap setiap keputusan prediksi sehingga meningkatkan interpretabilitas model. Hasil penelitian ini diharapkan dapat mendukung pengembangan sistem pendukung keputusan dalam pemilihan produk skincare yang lebih akurat dan transparan.
===================================================================================================================================
Selecting skincare products that are suitable for different skin types is a complex challenge due to variations in active ingredients and the biological characteristics of each skin type. The use of inappropriate skincare products may cause irritation, acne, or damage to the skin barrier function. Furthermore, the relationships among products, ingredients, product categories, and skin types are inherently complex, requiring an approach capable of representing these inter-entity relationships in a structured manner to produce more accurate and interpretable compatibility predictions. Therefore, this study employs a Knowledge Graph to model the semantic relationships among skincare products, product categories, ingredients, and skin types in the form of a directed graph and utilizes a Relational Graph Attention Network (RGAT) to learn graph representations by considering different relation types through the attention mechanism. The model input consists of node embeddings, edge indices, and edge types, while the output is the prediction of compatibility status (suitable or not suitable) between skincare products and specific skin types. The prediction results show that the model successfully identified the SKINTIFIC 5X Ceramide Barrier Moisture Gel product as suitable for oily skin based on the relationships among ingredients, product categories, and skin types represented in the Knowledge Graph. The evaluation results indicate that the proposed model achieved an accuracy of 84.28%, a precision of 85.05%, a recall of 83.18%, an F1-score of 84.10%, and an ROC AUC of 91.64%. In addition to predicting product compatibility, the model also provides attention weights that identify the most influential relationships for each prediction, thereby improving model interpretability. The results of this study are expected to support the development of a more accurate and transparent decision support system for selecting suitable skincare products.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Knowledge Graph, Relational Graph Attention Network, Skincare, Prediksi Kecocokan, Explainable Artificial Intelligence, Knowledge Graph, Relational Graph Attention Network, Skincare, Compatibility Prediction, Explainable Artificial Intelligence
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > Q Science (General) > Q325.78 Back propagation Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) R Medicine > RL Dermatology
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Mathematics > 44201-(S1) Undergraduate Thesis
Depositing User:	Juninda Mustofa
Date Deposited:	04 Aug 2026 01:20
Last Modified:	04 Aug 2026 01:20
URI:	http://repository.its.ac.id/id/eprint/142733

Actions (login required)

View Item