Pengenalan Perilaku Pengemudi Kendaraan Menggunakan Hybrid InceptionV3-BiLSTM

Shalihah, Frisalydia Nur (2025) Pengenalan Perilaku Pengemudi Kendaraan Menggunakan Hybrid InceptionV3-BiLSTM. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5002211104-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (7MB) | Request a copy

Abstract

Gangguan kognitif pada pengemudi merupakan salah satu penyebab utama kecelakaan lalu lintas. Untuk mengatasi hal ini, diperlukan sistem cerdas yang mampu mengenali perilaku pengemudi secara otomatis dan memberikan peringatan dini. Penelitian ini mengimplementasikan model hybrid berbasis Deep Learning, yaitu InceptionV3-Bidirectional Long Short-Term Memory (BiLSTM), yang menggabungkan keunggulan InceptionV3 dalam mengekstraksi fitur spasial dengan BiLSTM untuk memahami korelasi antar fitur. Dataset yang digunakan adalah data sekunder dari State Farm Distracted Driver Detection dan data primer yang mencakup 9 kategori perilaku pengemudi. Data diproses melalui tahap pre-processing, meliputi resizing, konversi tensor, dan normalisasi, kemudian dibagi menjadi data latih, data validasi, dan data uji. Model dilatih dengan data latih, dievaluasi menggunakan data validasi untuk mendapatkan model terbaik, dan diuji kinerjanya dengan data uji. Augmentasi dilakukan pada data latih berupa flipping horizontal, rotate, penyesuaian brightness, dan kontras. Model diterapkan dengan dua skenario, yaitu menggunakan 96 unit LSTM dan 64 unit LSTM, serta dilakukan variasi percobaan terhadap penggunaan dropout dan augmentasi. Evaluasi kinerja model hybrid InceptionV3-BiLSTM menunjukkan bahwa model dengan 96 unit LSTM, disertai dropout dan augmentasi, memberikan performa terbaik dengan nilai rata-rata recall sebesar 0.9531. Hasil ini menunjukkan potensi model dalam mendeteksi perilaku pengemudi secara akurat untuk mendukung sistem peringatan dini pada kendaraan.
======================================================================================================================================
Cognitive impairment in drivers is one of the main causes of traffic accidents. To address this issue, an intelligent system is needed that can automatically recognize driver behavior and provide early warnings. This study implements a hybrid model based on Deep Learning, namely InceptionV3-Bidirectional Long Short-Term Memory (BiLSTM), which combines the advantages of InceptionV3 in extracting spatial features with BiLSTM to understand the correlations between features. The dataset used consists of secondary data from the State Farm Distracted Driver Detection dataset and primary data covering nine categories of driver behavior. The data was processed through a pre-processing stage, including resizing, tensor conversion, and normalization, then divided into training data, validation data, and test data. The model was trained with the training data, evaluated using the validation data to obtain the best model, and tested for performance with the test data. Augmentation was performed on the training data in the form of horizontal flipping, rotation, brightness adjustment, and contrast. The model was applied in two scenarios, using 96 LSTM units and 64 LSTM units, and experiments were conducted on the use of dropout and augmentation. The performance evaluation of the InceptionV3-BiLSTM hybrid model shows that the model with 96 LSTM units, accompanied by dropout and augmentation, provides the best performance with an average recall value of 0.9531. These results demonstrate the model’s potential in accurately detecting driver behavior to support early warning systems in vehicles.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Pengenalan Perilaku Pengemudi, InceptionV3, BiLSTM, Dropout, Augmentasi, Driver Behavior Recognition, Augmentation
Subjects:	Q Science > QA Mathematics Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Mathematics > 44201-(S1) Undergraduate Thesis
Depositing User:	Frisalydia Nur Shalihah
Date Deposited:	29 Jul 2025 02:43
Last Modified:	29 Jul 2025 02:43
URI:	http://repository.its.ac.id/id/eprint/122424

Actions (login required)

View Item