Pengembangan Model Pendeteksi Kebohongan Berbasis Fotopletismografi untuk Meningkatkan Akurasi Deteksi Fisiologis Menggunakan Artificial Neural Network

Mizoguchi, Kentaro Mas'ud (2026) Pengembangan Model Pendeteksi Kebohongan Berbasis Fotopletismografi untuk Meningkatkan Akurasi Deteksi Fisiologis Menggunakan Artificial Neural Network. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6026251026-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (10MB) | Request a copy

Abstract

Penelitian ini mengembangkan model deteksi kebohongan (deception) berbasis sinyal fotopletismografi (PPG) menggunakan pendekatan machine learning dan deep learning. Data diperoleh dari 30 partisipan melalui sensor wearable Polar Verity Sense dan sensor gold standard Medlinket AM801, dengan skenario eksperimen meliputi Baseline, Game Theory, Concealed Information Test, dan Relevant Question Test. Sinyal PPG diproses melalui filtering, normalisasi, dan segmentasi, kemudian direpresentasikan dalam dua bentuk: fitur Heart Rate Variability (HRV) multidomain dan sinyal PPG mentah. Kedua representasi diklasifikasikan menggunakan enam algoritma, yaitu SVM, Random Forest, dan MLP untuk data HRV, serta 1D-CNN, CNN-LSTM, dan GRU untuk data raw PPG, dan dievaluasi melalui skenario Subject-Specific serta Leave-One-Subject-Out Cross-Validation (LOSOCV). Hasil menunjukkan model berbasis HRV (SVM dan Random Forest) mencapai akurasi tertinggi 0,922 pada skenario Subject-Specific, mengungguli model deep learning terbaik (1D-CNN, 0,781). Namun, seluruh model mengalami penurunan performa signifikan pada LOSOCV (akurasi 0,508-0,621), dengan MLP dan CNN-LSTM menunjukkan ketahanan generalisasi yang relatif lebih baik dibandingkan model yang unggul pada kondisi personalisasi. Temuan ini menunjukkan bahwa sinyal PPG berpotensi kuat sebagai indikator fisiologis kebohongan, namun kemampuan generalisasi model lintas individu masih menjadi tantangan yang perlu ditindaklanjuti sebelum diimplementasikan pada kondisi dunia nyata.
======================================================================================================================================
This research develops a deception detection model based on photoplethysmography (PPG) signals using machine learning and deep learning approaches. Data were collected from 30 participants using a wearable Polar Verity Sense sensor and a gold-standard Medlinket AM801 sensor, with an experimental protocol comprising Baseline, Game Theory, Concealed Information Test, and Relevant Question Test sessions. The PPG signals were processed through filtering, normalization, and segmentation, and subsequently represented in two forms: multidomain Heart Rate Variability (HRV) features and raw PPG signals. Both representations were classified using six algorithms, SVM, Random Forest, and MLP for HRV data, and 1D-CNN, CNN-LSTM, and GRU for raw PPG data, and evaluated under two scenarios: Subject-Specific and Leave-One-Subject-Out Cross-Validation (LOSOCV). The results show that HRV-based models (SVM and Random Forest) achieved the highest accuracy of 0.922 in the Subject-Specific scenario, outperforming the best deep learning model (1D-CNN, 0.781). However, all models exhibited a significant performance drop under LOSOCV (accuracy ranging from 0.508 to 0.621), with MLP and CNN-LSTM showing relatively better generalization resilience compared to the models that had performed best under the personalized condition. These findings indicate that PPG signals hold strong potential as a physiological indicator of deception, yet cross-subject generalization remains a key challenge that must be addressed before the model can be reliably implemented in real-world conditions.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Fotopletismografi, Deteksi Kebohongan, Sinyal Fisiologis, Artificial Neural Network, Variabilitas Denyut Jantung (HRV), Pemrosesan Sinyal Biomedis, Photoplethysmography, Lie Detection, Physiological Signals, Artificial Neural Network, Heart Rate Variability (HRV), Biomedical Signal Processing
Subjects:	Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information System > 59101-(S2) Master Thesis
Depositing User:	Kentaro Mas`ud Mizoguchi
Date Deposited:	27 Jul 2026 01:59
Last Modified:	27 Jul 2026 01:59
URI:	http://repository.its.ac.id/id/eprint/137624

Actions (login required)

View Item