research: The Classification of Mental States
My work as a PhD candidate focuses on developing innovative approaches to objective health monitoring. While methods like electroencephalography (EEG) have been used for over a century to measure brain activity, their application has historically been confined to clinical settings due to the cumbersome nature of traditional equipment. Building on the insights gained from studies in controlled environments, the emergence of wearable EEG devices offers a unique opportunity to bring this powerful technology into everyday life.
A primary focus of my research is the accurate classification of mental workload and stress, two closely related states that can have profound long-term health consequences. Too high levels of chronic mental workload and stress have been linked to an increased risk of cardiovascular disease and can contribute to deteriorating mental health, among others. Given the importance of early detection and intervention, accurate time-series classification is of utmost importance. Furthermore, the integration of active mental state management into computerized applications, like E-Learning, presents a major opportunity for educational advancement, among others. Contributing to this field, we conducted a systematic literature review on mental state classification using wearable EEG. Our findings highlighted the performance of multimodal time-series classification applications compared to their unimodal counterparts.
Building on this insight, my research combined wearable EEG with other physiological sensors, such as heart rate monitors, to create a more robust and accurate cognitive load detection system. My most recent research investigated differentiating levels of mental workload and task types across controlled, semi-controlled, and uncontrolled environments. These studies were designed to answer critical questions about the influence of noise on psychophysiological data and the reliability of wearable sensor systems in real-world conditions. A key aspect of this work was contributing to the scientific community through the release of anonymized, publicly available, high-quality datasets. Extensive data descriptors have been published, are accepted for publication, and planned, and some of the data is already accessible on Zenodo. Lastly, my final project used Bio-/Neurofeedback as an intervention to enhance learning activities and outcomes in E-Learning scenarios. This work seeked to create a tangible link between a user’s physiological state and their learning performance.
Zenodo dataset “UNIVERSE: UNobtrusIVE measuRement of mental workload and stress in uncontrolled environments”, available at https://zenodo.org/records/10371068, Zenodo dataset “Measuring and Quantifying Mental Workload and Stress in Everyday Situations, focusing on typical office activities”, available at https://zenodo.org/records/15681263, and Zenodo dataset “Cognitive Load Classification and Real-Time Intervention for Enhanced Vocabulary Learning”, available at https://zenodo.org/records/17350645.
I am always looking for motivated students and collaborators to expand on past and current projects, such as my studies “Measuring and Quantifying Mental Workload and Stress”, “Analysis of Synchronised Physiological Signals to Assess Mental Workload, Stress, Engagement, and Intervention-Effects on Small Groups”, and “Optimising E-Learning: The Wearable Sensors Opportunity to improve Learning Outcomes by Steering Mental Workload”. If you are interested in contributing to this research, please contact me to learn more about our study procedures and how you can get involved. I am also interested in supervising master's theses, particularly those that involve classification, prediction, or forecasting tasks on time-series data from wearable sensors. Projects that investigate the effects of data preprocessing on these tasks are equally welcome.
teaching
Before supervising master's theses, I jointly supervised with Sidratul Moontaha the master's project “Human Emotion and Activity Classification Using Brain Activity Sensors” (winter term 2021/2022)
This interdisciplinary project focused on the real-world distinction between work and relaxation using multimodal wearable sensors. The research spanned computer science, medicine, and psychology, integrating the processing of time-series medical data with an understanding of human affect and physiology. Students conducted an experiment to collect data from both a relaxed state and a state of mental workload using Empatica E4 and Muse S devices, which captured EEG, PPG, skin temperature, and GSR. The core of the work involved developing a robust signal processing pipeline, including techniques like the Savitzky-Golay filter and Common Spatial Pattern. For classification, the project compared various machine learning models, including Support Vector Machine and Random Forest, as well as Deep Convolutional Neural Networks. The results demonstrated a high level of accuracy, with the Random Forest model achieving 82.21% balanced accuracy in distinguishing between mental states. This project laid the groundwork for future research by creating a valuable use case for multimodal sensor data and a platform for further exploration, on which future master's theses built on.
Building on the experience gained in this project, I supervised the following three master's theses:
- Master's Thesis of Samik Real Enrı́quez: “Fighting Label Scarcity: Semi-automated human-in-the-loop label generation for automated stress and mental workload classification for uncontrolled environments” (jointly supervised with Sidratul Moontaha)
This thesis addressed the critical challenge of data scarcity in classifying mental workload and stress from wearable sensor data. It explored a semi-supervised learning approach to develop a human-in-the-loop system for generating labels in uncontrolled environments. By leveraging the principles of multimodality and deep learning, the research demonstrated a method to efficiently annotate large datasets produced by wearable devices. The findings established a pathway for creating reliable, labeled datasets for machine learning models, thereby making the detection of mental states in everyday scenarios more feasible and paving the way for future applications aimed at preventing stress-related health issues and conditions like epileptic seizures. - Master's Thesis of Sai Siddhant Gadamsetti: “Development and Validation of a Multimodal Data Acquisition and Analysis Platform for Assessing Workplace Stress in Group Settings”
This thesis laid the foundation for a comprehensive system to assess and manage workplace stress in group settings. The key accomplishments included the development of a Cognitive Load Induction Task (TLoadDback) to simulate real-world cognitive demands, the creation of a robust multimodal data acquisition system to record physiological responses, and the establishment of a sophisticated data analysis platform. A full-scale pilot study was successfully conducted to validate all components of the system. The project's findings provided a critical proof-of-concept for using objective, biometric measures to evaluate the effectiveness of workplace interventions, such as yoga, for stress reduction and well-being enhancement. - Master's Thesis of Ipsita Bhaduri: “Detection of mental workload and stress in realistic settings from multi-modal physiological signals”
This thesis successfully validated the use of multimodal wearable sensors for detecting and differentiating levels of mental workload and mild stress in both controlled laboratory and uncontrolled, real-world settings. The sensors used to collect physiological data were the Muse 2 and the Empatica E4, including EEG, heart rate variability, skin temperature, and galvanic skin response. A key achievement was the successful feature extraction and classification of these signals, demonstrating the potential of a low-cost, non-invasive setup to distinguish between different cognitive states.
publications
- Anders, C. & Arnrich, B., “Wearable electroencephalography and multi-modal mental state classification: A systematic literature review”, Computers in Biology and Medicine https://doi.org/10.1016/j.compbiomed.2022.106088
- Anders, C., Moontaha, S. & Arnrich, B., “Towards Multi-Modal Recordings in Daily Life: A Baseline Assessment of an Experimental Framework”, Proceedings 25th International Multiconference Information Society Pervasive Health and Smart Sensing (IS 2022) http://library.ijs.si/Stacks/Proceedings/InformationSociety/2022/IS2022_Volume-H%20-%20PHSS.pdf
- Anders, C., Curio, G., Arnrich, B. & Waterstraat, G., “Optimization of data pre-processing methods for time-series classification of electroencephalography data”, Network: Computation in Neural Systems https://doi.org/10.1080/0954898X.2023.2263083
- Anders, C., Gadamsetti, S. S., Steckhan, N. & Arnrich, B., “Load Induction then Simultaneous Relaxation: Insights from Multi-Modal Time-Series Data Measured with Low-Cost Wearable Sensors”, The Sixteenth International Conference on eHealth, Telemedicine, and Social Medicine (eTELEMED 2024) https://www.thinkmind.org/index.php?view=article\&articleid=etelemed_2024_1_50_40014
- Anders, C., Moontaha, S., Real, S., & Arnrich, B., “Unobtrusive measurement of cognitive load and physiological signals in uncontrolled environments”, Nature Sci Data 11, 1000 (2024). https://doi.org/10.1038/s41597-024-03738-7
- Anders, C., Bhaduri, I., & Arnrich, B., “Generalised machine learning models outperform personalised models for cognitive load classification in real-life settings”, Frontiers in Digital Health, Vol. 7, (2025). https://doi.org/10.3389/fdgth.2025.1650085
Plus (as of Oct. 2025) one manuscripts accepted, and two manuscripts in review.
personal
I love to cook, go for walks, travel and learn; DE (Muttersprache), EN (fluent), ES (avanzado)