Klasifikasi Tingkat Stres dan Analisis Mikroekspresi

Savero, Reynard Prastya (2025) Klasifikasi Tingkat Stres dan Analisis Mikroekspresi. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5023211042-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Mikroekspresi adalah ekspresi wajah spontan yang berlangsung sangat singkat (0,04-0,20 detik) dan mencerminkan emosi sesungguhnya tanpa dapat dikendalikan secara sadar. Stres merupakan respons psikologis dan fisiologis kompleks tubuh terhadap tekanan yang diatur oleh sistem neuroendokrin dan berdampak signifikan pada kesehatan fisik maupun mental. Deteksi dan klasifikasi kedua aspek ini memiliki potensi besar dalam bidang psikologi, kesehatan mental, dan keamanan. Namun, metode deteksi manual yang ada saat ini memiliki keterbatasan dalam hal efisiensi, akurasi, dan subjektivitas, sementara sistem otomatis yang tersedia memerlukan perangkat komputasi berbiaya tinggi dengan portabilitas terbatas. Penelitian ini mengembangkan sistem terintegrasi berbasis machine learning untuk deteksi mikroekspresi dan klasifikasi tingkat stres menggunakan platform Raspberry Pi 5 yang dilengkapi kamera Raspberry Pi Camera V3. Sistem memanfaatkan perangkat lunak OpenFace 2.0 untuk mengekstraksi 18 fitur Action Unit (AU) dari video subjek. Data AU kemudian diproses melalui dua modul klasifikasi. Modul pertama menggunakan algoritma Support Vector Machine (SVM) untuk mengklasifikasikan mikroekspresi ke dalam enam kategori emosi dasar yaitu bahagia, marah, jijik, takut, sedih, dan terkejut. Modul kedua menggunakan Artificial Neural Network (ANN) untuk mengklasifikasikan tingkat stres menjadi tiga kategori: rendah, sedang, dan tinggi. Protokol induksi stres menggunakan Trier Social Stress Test yang mencakup sesi baseline (membaca teks netral), stres sedang (presentasi dan perhitungan matematika), dan stres tinggi (kalkulasi serial mundur dan simulasi Montreal Imaging Stress Task). Pengujian dilakukan pada 71 subjek dengan total 213 sampel data yang terdistribusi seimbang. Sistem SVM mencapai akurasi keseluruhan 82,0% dengan F1-score 0,81 untuk klasifikasi mikroekspresi. Performa per kategori menunjukkan variasi signifikan: emosi marah mencapai akurasi tertinggi (90%) dengan F1-score sempurna (0,9000), diikuti bahagia (78%), takut (75%), jijik (70%), terkejut (57%), dan sedih sebagai kategori dengan performa terendah (57%). Sistem ANN untuk klasifikasi stres mencapai akurasi keseluruhan 84,98% dengan distribusi yaitu stres tinggi memiliki akurasi tertinggi (88,7%), stres rendah (85,9%), dan stres sedang (80,3%).
=================================================================================================================================
Micro-expressions are brief, involuntary facial expressions lasting between 0.04 and 0.20 seconds that reveal genuine emotions beyond conscious control. Stress, on the other hand, is a complex psychological and physiological response to external pressure, regulated by neuroendocrine systems and significantly affecting both physical and mental well-being. Accurate detection and classification of micro-expressions and stress levels hold great promise in fields such as psychology, mental health, and security. However, existing manual detection methods suffer from inefficiency, subjectivity, and limited accuracy, while current automated systems often rely on high-cost computing devices that lack portability. This research proposes an integrated artificial intelligence based system for simultaneous micro-expression detection and stress level classification, implemented on a Raspberry Pi 5 platform equipped with a Raspberry Pi Camera V3. The system utilizes OpenFace 2.0 to extract 18 Action Unit (AU) features from subject video recordings. The extracted AU data is processed through two classification modules: the first uses a Support Vector Machine (SVM) algorithm to classify micro-expressions into six basic emotions happiness, anger, disgust, fear, sadness, and surprise. The second employs an Artificial Neural Network (ANN) to classify stress levels into three categories: low, medium, and high. Stress induction was carried out using the Trier Social Stress Test (TSST), consisting of a baseline session (neutral text reading), a moderate stress session (presentation and arithmetic tasks), and a high-stress session (serial subtraction and a simulated Montreal Imaging Stress Task). The system was tested on 71 subjects, yielding 213 data samples evenly distributed across the categories. The SVM module achieved an overall accuracy of 82.0% with an F1-score of 0.81 for micro-expression classification. Emotion-specific performance varied, with anger achieving the highest accuracy (90%) and a perfect F1-score (0.9000), followed by happiness (78%), fear (75%), disgust (70%), and surprise and sadness both at 57%. The ANN module for stress classification reached an overall accuracy of 84.98%, with category-specific accuracies of 88.7% for high stress, 85.9% for low stress, and 80.3% for medium stress.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Kata kunci: Mikroekspresi, Raspberry Pi, Action Unit, Support Vector Machine, OpenFace 2.0, Klasifikasi Stres, Artificial Neural Network, Trier Social Stress Test ============================================================================================================================== Keywords: Micro-expression, Raspberry Pi, Action Unit, Support Vector Machine, OpenFace 2.0, Stress Classification, Artificial Neural Network, Trier Social Stress Test
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7882.P3 Pattern recognition systems
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Biomedical Engineering > 11410-(S1) Undergraduate Thesis
Depositing User:	Reynard Prastya Savero
Date Deposited:	04 Aug 2025 03:34
Last Modified:	04 Aug 2025 03:34
URI:	http://repository.its.ac.id/id/eprint/126604

Actions (login required)

View Item