Fiddiyansyah, Rizka (2025) REKOMENDASI PRODUK SKINCARE PRIA BERDASARKAN KLASIFIKASI KULIT WAJAH BERBASIS FOTO MENGGUNAKAN CNN PADA WEBSITE BROMEN.ID. Undergraduate thesis, UPN Veteran Jawa Timur.
![]() |
Text (Cover)
21082010052.-cover.pdf Download (1MB) |
![]() |
Text (Bab 1)
21082010052.-bab1.pdf Download (435kB) |
![]() |
Text (Bab 2)
21082010052.-bab2.pdf Restricted to Repository staff only until 25 July 2027. Download (1MB) |
![]() |
Text (Bab 3)
21082010052.-bab3.pdf Restricted to Repository staff only until 25 July 2027. Download (324kB) |
![]() |
Text (Bab 4)
21082010052.-bab4.pdf Restricted to Repository staff only until 25 July 2027. Download (6MB) |
![]() |
Text (Bab 5)
21082010052.-bab5.pdf Download (225kB) |
![]() |
Text (Daftar pustaka)
21082010052.-daftar pustaka.pdf Download (204kB) |
![]() |
Text (Lampiran)
21082010052.-lampiran.pdf Restricted to Repository staff only until 25 July 2027. Download (1MB) |
Abstract
Men's awareness of the importance of healthy facial skin continues to increase, but are often faced with difficulties in choosing skincare products that suit their unique skin characteristics, such as oily skin tendencies and potential irritation from shaving. Conventional recommendation methods, such as the questionnaire currently used on the Bromen.id website, are considered less accurate and prone to bias. This study aims to develop a men's skincare product recommendation system through photo-based facial skin type classification using Convolutional Neural Network (CNN), find the most optimal model, and integrate it into the Bromen.id website. The methodology used adopts the CRISP-DM framework, including facial image data collection (including data synthesis to overcome class imbalance and enrich diversity), data preparation (division into training, validation, and testing sets with various ratios, as well as augmentation for diversity), and model building using traditional CNN as well as Transfer Learning (MobileNet, VGG-16, ResNet-50). The model training process was performed in two scenarios: without callbacks with a full 100 epochs to see the baseline performance of the model, and with EarlyStopping and ReduceLROnPlateau callbacks for convergence optimisation and overfitting prevention. A comprehensive evaluation was conducted using accuracy, loss, precision, recall, F1-score, and confusion matrix metrics. The results showed that the MobileNet model with a data split ratio of 80:10:10 proved to be the most optimal with the implementation of callbacks, achieving an accuracy of 0.99 on training data and 0.96 on test data, as well as excellent performance on the validation set. Although the 'Normal' class was still a challenge with the highest percentage prediction error, the model overall showed the best stability and generalisability. The integration of the model into the Bromen.id website was successful, providing personalised and accurate product recommendations based on user facial photo analysis.
Item Type: | Thesis (Undergraduate) | ||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Contributors: |
|
||||||||||||
Subjects: | T Technology > T Technology (General) | ||||||||||||
Divisions: | Faculty of Computer Science > Departemen of Information Systems | ||||||||||||
Depositing User: | Rizka Fiddiyansyah | ||||||||||||
Date Deposited: | 25 Jul 2025 09:17 | ||||||||||||
Last Modified: | 25 Jul 2025 09:17 | ||||||||||||
URI: | https://repository.upnjatim.ac.id/id/eprint/41058 |
Actions (login required)
![]() |
View Item |