Rabbani, Zufar Abdullah (2026) KLASIFIKASI AKTIVITAS DISTRAKSI PENGEMUDI BERDASARKAN CITRA MENGGUNAKAN EFFICIENTNETB0 PADA KONDISI PENCAHAYAAN BERAGAM DENGAN CHANNEL ATTENTION DAN AUGMENTASI DATA. Undergraduate thesis, UPN Veteran Jawa Timur.
|
Text (Cover)
22083010080-cover.pdf Download (1MB) | Preview |
|
|
Text (Bab 1)
22083010080-bab1.pdf Download (249kB) | Preview |
|
![[img]](https://repository.upnjatim.ac.id/style/images/fileicons/text.png) |
Text (Bab 2)
22083010080-bab2.pdf Restricted to Repository staff only until 7 July 2029. Download (1MB) |
|
|
Text (Bab 3)
22083010080-bab3.pdf Restricted to Repository staff only until 7 July 2029. Download (811kB) |
|
|
Text (Bab 4)
22083010080-bab4.pdf Restricted to Repository staff only until 7 July 2029. Download (2MB) |
|
|
Text (Bab 5)
22083010080-bab5.pdf Download (227kB) | Preview |
|
|
Text (Daftar pustaka)
22083010080-daftarpustaka.pdf Download (203kB) | Preview |
|
|
Text (Lampiran)
22083010080-lampiran.pdf Restricted to Repository staff only until 7 July 2029. Download (337kB) |
Abstract
Distracted driving is a major contributing factor to traffic accidents in various countries. In Indonesia, data from the Indonesian National Police Traffic Corps (Korlantas Polri) recorded 94,600 accidents, most of which were caused by distracted driving. Efficient Convolutional Neural Network (CNN) architectures such as EfficientNetB0 have been used by previous studies to classify distracted driving activities based on imagery in an effort to reduce the risk of accidents due to human error, and have been shown to produce high accuracy with relatively low computational resources. EfficientNetB0 offers efficiency through its compound scaling design, which simultaneously optimizes accuracy and computational consumption. Therefore, this study uses EfficientNetB0 because in distracted driver classification, real-time scenarios are crucial. The addition of Channel Attention (CA) and data augmentation to EfficientNetB0 is also used because it has been proven to improve accuracy by focusing on important features and making the model more robust without sacrificing model efficiency. Previous distracted driver classification studies have only used public datasets with relatively similar lighting conditions, thus not reflecting real-world conditions with varying lighting. Previous studies have also shown that lighting affects the performance of CNN architectures. This study fills this gap by classifying distracted drivers under varying lighting conditions using EfficientNetB0 with attention channels and data augmentation. The results of this study indicate that adding attention channels to the EfficientNetB0 backbone tends to worsen the model's accuracy, while adding data augmentation further improves the model's accuracy for classification under varying lighting conditions. This is indicated by the pure EfficientNetB0 backbone accuracy value of 67% in all lighting conditions, when added with augmentation techniques (without CA) the accuracy value becomes 88%, but when added with attention channels (without augmentation) the accuracy value becomes 59%. This shows that the EfficientNetB0 model with data augmentation (without CA) is the best model with an accuracy value of 88% for dark conditions, 86% for medium conditions, and 90% for bright conditions.
| Item Type: | Thesis (Undergraduate) | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Contributors: |
|
||||||||||||
| Subjects: | Q Science > QA Mathematics > QA76.6 Computer Programming | ||||||||||||
| Divisions: | Faculty of Computer Science > Departemen of Data Science | ||||||||||||
| Depositing User: | Zufar Rabbani | ||||||||||||
| Date Deposited: | 07 Jul 2026 07:03 | ||||||||||||
| Last Modified: | 07 Jul 2026 07:35 | ||||||||||||
| URI: | https://repository.upnjatim.ac.id/id/eprint/54746 |
Actions (login required)
 |
View Item |