Biometrics and Machine Learning

Author: Lauren Devine

The use of biometrics in research has grown in popularity due to its novel ability to investigate subconscious human behavior. Analyzing various bodily functions can reveal otherwise unknown features related to attention, emotion, cognition, and physiological arousal. Popular tools at the forefront of this discovery include eye trackers, electroencephalograms (EEGs), and galvanic skin response sensors. Respectively, these devices can capture bodily activities through the forms of ocular movement, neural waves, and perspiration.

These devices have the power to capture immense amounts of data for interpretation. However, this data is often convoluted, dense, and difficult to interpret. Using sheer manpower to extract and comprehend these vast scales of information is often futile. However, breakthroughs in machine learning (ML) and deep learning (DL) have created breakthroughs in data analysis. If developed appropriately, these complex algorithms can filter through noisy datasets to find key results and areas of interest (Saedi 2021).

Current ML models utilized in biometrics research dominate in the field of classification. The objective of classification is to use input (or training) data to create algorithms that classify new data into the appropriate label. These are supervised algorithms that must “learn” how to sort the data. For example, an ML model based on image comprehension could be trained on sample data to “learn” what a dog looks like. Then, when pictures of various animals are shown, the images can be classified into “dog” and “not dog” categories.

There are many types of standard algorithms utilized when creating a machine learning model. When using these models to classify biometric data, researchers tend to choose models that have the highest accuracy percentages. Different types of sensors and research goals require different interpretations, so the algorithms differ depending on the study. The most commonly used models include Decision Tree, Random Forest, Support Vector Machine, and K-Nearest Neighbor (Rosidi 2023).

The intersection of biometrics and machine learning has the power to reveal novel insights about human behavior. Though a relatively new field, immense progress has been made in analyzing brain activity, skin conductivity, and eye motions.

Electroencephalography

Electroencephalography (EEG) is the most popular method to study the brain’s electrical activity. This non-invasive method tracks dynamic movements in the human brain by measuring voltage fluctuations from the scalp. This neuroimaging technique has become a tool in many neuroscientific breakthroughs, including sleep analysis, behavioral studies, clinical diagnoses, and brain-computer interface systems. However, raw EEG signals are complex and noisy, requiring many steps of data refinement (Saedi 2021).

Machine learning and deep learning methods have quickly grown as the primary methods to decode EEG results and create meaningful conclusions. These tools can quickly pick through extraneous data (i.e. eye blinking, movement, and respiration) to find applicable brain wave data and make appropriate classifications (Saedi 2021).

ML and DL algorithms have been used as classification instruments in many cognitive and clinical studies utilizing EEGs. Recent studies have supported the fields of psychology, medicine, and diagnostics:

• A Support Vector Machine (SVM) was used to detect emotions based on raw EEG signals. This program achieved 95.5% accuracy for classifying four emotions (happiness, fear, sadness, and relaxation) (Khare et al., 2020).

• EEG sleep signals have been extracted and trained on Random Forest ML algorithms to classify sleep stages. This program has resulted in 97.8% accuracy in predicting sleep stages (Sagar, Desai).

• Higher-order spectra is a powerful method to extract nonlinear EEG signal features of Parkinson’s Disease. This data was fed through an SVM to differentiate between affected and unaffected patients. This program reached a diagnostic accuracy of 99.62% (Yuvaraj et al., 2018).

Galvanic Skin Response

Galvanic Skin Response (GSR) sensors measure the changes in the variable electrical properties in skin due to perspiration. This method involves placing sensors onto the skin that monitor sweat gland activity. GSR provides insight into emotional arousal. When an individual has exposure to an arousing stimulus, the heart beats faster, pulse increases, sweat secretion increases, and GSR will increase as well (Satti et al., 2021). GSR is a valuable tool that leverages data based on subconscious bodily functions, offering insights into an individual’s physiological and psychological activity.

Advancements in machine learning have provided opportunities for more complex analysis of GSR signals. Historically, GSR data has been biased, user-dependent, and convoluted due to the nature of skin precipitation differences by individual. Machine learning methods can sift through these datasets to create classifications of arousal (Georgas et al., 2025).

The introduction of machine learning into GSR studies has had meaningful results in the fields of psychology and healthcare:

• A majority voting technique utilizing 7 ML algorithms detected stress and relaxation levels in games. This technique yielded a 63.39% level of accuracy in predicting mental states for subjects (Satti et al., 2021).

• Sensors were used to study the impact of COVID-19 rapid tests on patients' stress levels. A specialized SVM ML tool found with 70% accuracy that GSR notably increased during these rapid tests, indicating increased anxiety (Georgas et al., 2025).

• A study on anxiety paired GSR signals with K-nearest neighbor (KNN) algorithms to classify anxiety levels. The program successfully detected anxiety in subjects with a statistical accuracy of 96.9% (Al-Nafjan, Aldayel, 2024).

Eye Tracking

Eye tracking technology monitors subjects' eye movements. This process utilizes near-infrared technology along with high-resolution cameras to detect large and subtle motions. Key metrics of tracked ocular activity include gaze, fixations, and eye movements called saccades. Eye motion is tightly linked to visual attention and human interaction. Though eye tracking alone can’t explain how people perceive a stimulus, the biometric data can reveal a wealth of information about human behavior and cognition (Lim et al., 2022).

Machine learning has revolutionized the understanding of eye-tracking data. Innovations in this field integrate eye-tracking devices with machine learning. This seamless combination can quickly filter through the various ocular measurements to find meaningful results (Lim et al., 2022).

Recent studies that highlight the use of machine learning and eye tracking have led to discoveries in the areas of childhood development, psychology, and diagnostics:

• SVM and DL models have successfully classified toddlers' ages using eye tracking. This study identified meaningful features that distinguish between age groups, including gaze patterns and fixation metrics (Dalrymple 2019).

• Changes in pupil diameter and eyeball movements, paired with decision tree algorithms, have classified whether a person is lying with 95% accuracy (Labibah et al., 2018).

• Individuals were tested for autism using eye tracking as they watched a video and were asked to recognize the emotions shown. Utilizing random forest algorithms alongside this data achieved an 86% accuracy in diagnosis (Jiang et al., 2019).

Future of Biometrics and Machine Learning

The use of ML to decipher biometric information has been crucial to the progression of human behavior research. Though these biometric devices yield powerful results alone, when the devices are paired together, they create more comprehensive insights into the unconscious mind. These clinical studies may unlock new ways to treat patients in need:

• A study utilizing EEG and GSR devices paired with logistical regression could distinguish methamphetamine-dependent patients from healthy subjects with an accuracy of 90.68% (Ding et al., 2020).

• A combination of portable EEG, eye tracking, and GSR devices used logistical regression to classify depression patients from healthy controls with an accuracy of 79.63% (Ding et al., 2019).

Loyola’s biometrics lab has access to many tools that are at the forefront of biometrics research. Researchers have access to GSR devices, eye trackers, and Facial Action Coding Systems. With the support of these tools, the lab is continuously pushing innovation in the study of human behavior.

Links for Further Exploration:

• Machine learning model basics: https://www.stratascratch.com/blog/a-comprehensive-overview-of-3-popular-machine-learning-models/

• Literature review on EEG signals with ML: https://pmc.ncbi.nlm.nih.gov/articles/PMC8615531/#sec4-brainsci-11-01525

• Literature review on eye-tracking data with ML: https://www.frontiersin.org/journals/neurorobotics/articles/10.3389/fnbot.2021.796895/full#B17

• Utilizing eye tracking to classify age: https://pubmed.ncbi.nlm.nih.gov/31000762/

• Utilizing various biometric tools to diagnose depression: https://www.sciencedirect.com/science/article/abs/pii/S0165032718330064.

References 

1. Al-Nafjan, Abeer, and Mashael Aldayel. "Anxiety detection system based on galvanic skin response signals." Applied Sciences 14.23 (2024): 10788.

2. Dalrymple, Kirsten A., et al. "Machine learning accurately classifies age of toddlers based on eye tracking." Scientific reports 9.1 (2019): 6255.

3. Ding, Xinfang et al. “Using machine-learning approach to distinguish patients with methamphetamine dependence from healthy subjects in a virtual reality environment.” Brain and behavior vol. 10,11 (2020): e01814. doi:10.1002/brb3.1814

4. Ding, Xinfang, et al. "Classifying major depression patients and healthy controls using EEG, eye tracking and galvanic skin response data." Journal of affective Disorders 251 (2019): 156-161.

5. F. A. Satti, M. Hussain, J. Hussain, T. -S. Kim, S. Lee and T. Chung, "User Stress Modeling through Galvanic Skin Response," 2021 15th International Conference on Ubiquitous Information Management and Communication (IMCOM), Seoul, Korea (South), 2021, pp. 1-6

6.  Georgas, Antonios, et al. "Stress Monitoring in Pandemic Screening: Insights from GSR Sensor and Machine Learning Analysis." Biosensors 15.1 (2025): 14.

7.  Jiang, Ming, et al. "Classifying individuals with ASD through facial emotion recognition and eye-tracking." 2019 41st Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC). IEEE, 2019.

8.  Khare, Smith K., Varun Bajaj, and Ganesh Ram Sinha. "Adaptive tunable Q wavelet transform-based emotion identification." IEEE transactions on instrumentation and measurement 69.12 (2020): 9609-9617.

9. Labibah, Zuhrah, Muhammad Nasrun, and Casi Setianingsih. "Lie detector with the analysis of the change of diameter pupil and the eye movement use method Gabor wavelet transform and decision tree." 2018 IEEE International Conference on Internet of Things and Intelligence System (IOTAIS). IEEE, 2018.

10. Lim, Jia Zheng, James Mountstephens, and Jason Teo. "Eye-tracking feature extraction for biometric machine learning." Frontiers in neurorobotics 15 (2022): 796895.

11. Rosidi, Nathan. “A Comprehensive Overview of 3 Popular Machine Learning Models.” (2023), StrataScratch, www.stratascratch.com/blog/a-comprehensive-overview-of-3-popular-machine-learning-models.

12.  Saeidi, Maham et al. “Neural Decoding of EEG Signals with Machine Learning: A Systematic Review.” Brain sciences vol. 11,11 1525. 18 Nov. 2021, doi:10.3390/brainsci11111525

13.  Santaji, Sagar, and Veena Desai. "Analysis of EEG signal to classify sleep stages using machine learning." Sleep and Vigilance 4.2 (2020): 145-152.

14.  Yuvaraj, Rajamanickam, U. Rajendra Acharya, and Yuki Hagiwara. "A novel Parkinson’s Disease Diagnosis Index using higher-order spectra features in EEG signals." Neural Computing and Applications 30 (2018): 1225-1235.