Bi-GRU Based Deception Detection using EEG Signals
Danilo Avola, Muhammad Yasir Bilal, Emad Emam, Cristina Lakasz · 2025
Read Bi-GRU Based Deception Detection using EEG Signals by Danilo Avola, Muhammad Yasir Bilal, Emad Emam, Cristina Lakasz (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-13718.txt.
Introduction
Danilo Avola, Muhammad Bilal, Emad Emam, Cristina Lakasz, Daniele Pannone, and Amedeo Ranaldi catch lies from brainwaves with a bidirectional recurrent network. Their paper from twenty twenty-five trains on the Bag-of-Lies dataset, a multimodal corpus of casual, natural deception. The classifier reads EEG signals only. A Bi-GRU network scans each signal forward and backward, catching patterns that one-way passes miss.
Test accuracy lands at ninety-seven percent, with precision, recall, and F1-scores high for both truth and lies. The authors credit bidirectional time modeling for the strength. Real-time deception screening is floated as a use, alongside richer architectures later. Security, psychology, and forensics all watch this space, and the result is a strong mark on a public benchmark.
Catalog record
| CREATOR | Danilo Avola, Muhammad Yasir Bilal, Emad Emam, Cristina Lakasz |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 32 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2507-13718
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Bi-GRU Based Deception Detection using EEG Signals Danilo Avola, Muhammad Yasir Bilal, Emad Emam, Cristina Lakasz Abstract Deception detection is a significant challenge in fields such as security, psychology, and forensics. This study presents a deep learning approach for classifying deceptive and truthful behavior using ElectroEncephaloGram (EEG) signals from the Bag-of-Lies dataset, a multimodal corpus designed for naturalistic, casual deception scenarios. A Bidirectional Gated Recurrent Unit (Bi-GRU) neural network was trained to perform binary classification of EEG samples. The model achieved a test accuracy of 97%, along with high precision, recall, and F1-scores across both classes. These results demonstrate the effectiveness of using bidirectional temporal modeling for EEG-based deception detection and suggest potential for real-time applications and future exploration of advanced neural architectures. Index terms— EEG, Deception Detection, Deep Learning, Bi-GRU 1 Introduction Computer vision and machine learning have emerged as pivotal technologies enabling a wide range of applications, including robust feature extraction [ 6 , 8 ] , person re-identification in surveillance systems [ 4 , 13 ] , and intelligent drone navigation [ 5 , 11 ] . Moreover, recent advancements extend their impact to more sensitive domains such as lie detection [ 7 , 3 ] , demonstrating their growing role in addressing complex real-world challenges. Deception detection refers to the process of identifying statements or actions that are designed to conceal the truth. Lying is a common habitual practice, intrinsic in people daily life. It is characterized by an intentional effort to mislead others for personal benefit or gain and is a behavior that can manifest in various forms, such as fabrications, omissions, and misrepresentations. Fabrications involve the creation of false information or stories, omissions refer to the deliberate exclusion of certain truths, and misrepresentations involve presenting information in a way that distorts the truth or creates a false impression. The spectrum of deception is thus broad, ranging from harmless white lies to serious threats that pose significant morality and security risks to society, with far-reaching implications, affecting interpersonal relationships, organizational dynamics, and even national security. For these reasons, the importance of deception detection cannot be overstated, as it plays a crucial role in various aspects of society, including law enforcement and corporate governance. By accurately identifying deceptive behavior, we can foster transparency, accountability, and trust in our interactions. In recent studies [ 15 ] , the methods for detecting deception have been broadly divided into two main categories: non-contact based and contact based methods. The first primarily focuses on the extraction of biological and physiological data that measure activity changes within the body, such as heart rate, blood pressure, respiratory patterns and, most importantly for our study, bioelectrical signals. The second group includes methods that rely on acoustic, visual, thermal data, and verbal communication. Acoustic data can provide insights into changes in a person’s voice, while verbal data can reveal deceptive cues in the content of a person’s speech. Thermal data can show physiological changes such as increased body temperature, and visual data, such as facial expressions or body language, can provide clues about whether a person is telling the truth or attempting to deceive [ 14 ] . By exploiting modern deep learning techniques, this paper introduces a novel application of a Bidirectional Gated Recurrent Unit (Bi-GRU) neural network architecture for classifying truthful and deceptive behavior based solely on EEG signals, specifically focusing on the temporal dynamics of brain activity. The following are the contributions of our work: • Exclusive Focus on Spontaneous Deception: Utilizes the EEG portion of the Bag-of-Lies dataset, which captures spontaneous, self-initiated deception in realistic scenarios, unlike many prior datasets relying on forced or artificial deception; • Robust pre-processing and augmentation pipeline: Proposes a structured pre-processing method that includes filtering, overlapping window segmentation, class balancing via undersampling, and data augmentation by Gaussian noise to address dataset limitations; • High Classification Accuracy and Generalization: Achieves a 97% accuracy on the EEG classification task, outperforming several state-of-the-art models applied to the same dataset, including Random Forest, Bi-LSTM, and CNN-based approaches. The rest of the paper is organized as follows. Section 2 , the current state-of-the-art methods in deception detection through EEG are discussed. Section 3 presents the designed method, together with the description of the pre-processing operations performed on the dataset. Section 4 report the conducted experiments and the results obtained. Finally, Section 5 concludes the paper. 2 Related work The ability to detect lies is of critical importance in legal and security settings, for this reason research on deception detection has been extensively conducted. With advances in technology, particularly in the field of neuroscience and artificial intelligence, new methods have emerged to improve the accuracy and reliability of lie detection. We will particularly focus on EEG applications in this field. Much of the work related to deception detection using EEG data is focused specifically on P300 waves. These are components of the Event-Related Potential (ERP) characterized by a positive deflection in voltage that occurs approximately 300 milliseconds after the presentation of a stimulus and is often more pronounced when a subject recognizes the stimulus as familiar or meaningful. We now discuss briefly some works that inspired our study, but that focus on less controlled experiments, closer to real life scenarios and that do not leverage P300 waves. Authors in [ 9 ] present the Bag-Of-Lies dataset a pivotal multimodal dataset in this field. It integrates video, audio, EEG, and gaze data, collected from 35 unique subjects in a realistic setting. It provides a balanced set of 325 annotated data points divided almost equally between truthful and deceptive responses. The video data is extracted using Local Binary Patterns (LBP) and analyzed thru SVM, Random Forest, and MLP. Extracted features from the audio data, such as spectral properties and mel-frequency cepstral coefficients (MFCCs), are classified using Random Forest and KNN. EEG data is analyzed through sub-sampling using a sliding window approach, with features fed into a CNN architecture and Random Forest for classification. Finally, gaze data features are calculated included fixations, eye blinks, and pupil size, which are then used to construct a feature vector for machine learning analysis. Individual modal analyses showed varied effectiveness, the EEG data provided moderate results, with the best classification accuracy using Random Forest achieving 58.71%. Audio data achieved 56.20% accuracy with Random Forest. Similarly to video, audio data provided modest results with the best performance shown by KNN at 56.22% accuracy. Gaze data performed better than all other modalities alone, with Random Forest achieving the highest accuracy at 61.70%. Performing multimodal fusion classification, particularly combining gaze, video, and audio data, yielded the highest improvements in detection accuracy, 64.69%. Lastly, the combination of all four modalities (Video, Audio, EEG, and Gaze) resulted in the highest accuracy of 66.17% highlighting the benefit of a comprehensive analytical approach. Furthermore, the EEG data of the Bag-Of-Lies dataset is the one used in this study. For this reason, part of this dataset, will be explained in greater detail in the following chapters. The work presented in [ 12 ] compares a deep convolutional neural network (DCNN) designed for detecting deception on three multimodal databases: Bag-of-Lies (BoL), Real-Life (RL) trail, and Miami University Deception Detection (MU3D). Visual, vocal, and EEG signals of BoL