Search

Stress recognition, particularly using machine learning (ML) with physiological data such as heart rate variability (HRV), holds promise for mental health interventions. However, limited datasets in affective computing and healthcare research can lead to inaccurate conclusions regarding the ML model performance. This study employed supervised learning algorithms to classify stress and relaxation states using HRV measures. To account for limitations associated with small datasets, robust strategies were implemented based on methodological recommendations for ML with a limited dataset, including data segmentation, feature selection, and model evaluation. Our findings highlight that the random forest model achieved the best performance in distinguishing stress from non-stress states. Notably, it showed higher performance in identifying stress from relaxation (F1-score: 86.3%) compared to neutral states (F1-score: 65.8%). Additionally, the model demonstrated generalizability when tested on independent secondary datasets, showcasing its ability to distinguish between stress and relaxation states. While our performance metrics might be lower than some previous studies, this likely reflects our focus on robust methodologies to enhance the generalizability and interpretability of ML models, which are crucial for real-world applications with limited datasets.

The perception of sound greatly impacts users’ emotional states, expectations, affective relationships with products, and purchase decisions. Consequently, assessing the perceived quality of sounds through jury testing is crucial in product design. However, the subjective nature of jurors’ responses may limit the accuracy and reliability of jury test outcomes. This research explores the utility of facial expression analysis in jury testing to enhance response reliability and mitigate subjectivity. Some quantitative indicators allow the research hypothesis to be validated, such as the correlation between jurors’ emotional responses and valence values, the accuracy of jury tests, and the disparities between jurors’ questionnaire responses and the emotions measured by FER (facial expression recognition). Specifically, analysis of attention levels during different statuses reveals a discernible decrease in attention levels, with 70 percent of jurors exhibiting reduced attention levels in the ‘distracted’ state and 62 percent in the ‘heavy-eyed’ state. On the other hand, regression analysis shows that the correlation between jurors’ valence and their choices in the jury test increases when considering the data where the jurors are attentive. The correlation highlights the potential of facial expression analysis as a reliable tool for assessing juror engagement. The findings suggest that integrating facial expression recognition can enhance the accuracy of jury testing in product design by providing a more dependable assessment of user responses and deeper insights into participants’ reactions to auditory stimuli.

Facial expression recognition (FER) poses a complex challenge due to diverse factors such as facial morphology variations, lighting conditions, and cultural nuances in emotion representation. To address these hurdles, specific FER algorithms leverage advanced data analysis for inferring emotional states from facial expressions. In this study, we introduce a universal validation methodology assessing any FER algorithm’s performance through a web application where subjects respond to emotive images. We present the labelled data database, FeelPix, generated from facial landmark coordinates during FER algorithm validation. FeelPix is available to train and test generic FER algorithms, accurately identifying users’ facial expressions. A testing algorithm classifies emotions based on FeelPix data, ensuring its reliability. Designed as a computationally lightweight solution, it finds applications in online systems. Our contribution improves facial expression recognition, enabling the identification and interpretation of emotions associated with facial expressions, offering profound insights into individuals’ emotional reactions. This contribution has implications for healthcare, security, human-computer interaction, and entertainment.

Most studies have demonstrated that EEG can be applied to emotion recognition. In the process of EEG-based emotion recognition, real-time is an important feature. In this paper, the real-time problem of emotion recognition based on EEG is explained and analyzed. Secondly, the short time window length and attention mechanisms are designed on EEG signals to follow emotion change over time. Then, long short-term memory with the additive attention mechanism is used for emotion recognition, due to timely emotion updates, and the model is applied to the SEED and SEED-IV datasets to verify the feasibility of real-time emotion recognition. The results show that the model performs relatively well in terms of real-time performance, with accuracy rates of 85.40% and 74.26% on SEED and SEED-IV, but the accuracy rate has not reached the ideal state due to data labeling and other losses in the pursuit of real-time performance.

This paper presents novel preliminary research that investigates the relationship between the flow of a group of jazz musicians, quantified through multi-person pose synchronization, and their collective emotions. We have developed a real-time software to calculate the physical synchronicity of team members by tracking the difference in arm, leg, and head movements using Lightweight OpenPose. We employ facial expression recognition to evaluate the musicians’ collective emotions. Through correlation and regression analysis, we establish that higher levels of synchronized body and head movements correspond to lower levels of disgust, anger, sadness, and higher levels of joy among the musicians. Furthermore, we utilize 1-D CNNs to predict the collective emotions of the musicians. The model leverages 17 body synchrony keypoint vectors as features, resulting in a training accuracy of 61.47% and a test accuracy of 66.17%.

A core endeavour in current affective computing and social signal processing research is the construction of datasets embedding suitable ground truths to foster machine learning methods. This practice brings up hitherto overlooked intricacies. In this paper, we consider causal factors potentially arising when human raters evaluate the affect fluctuations of subjects involved in dyadic interactions and subsequently categorise them in terms of social participation traits. To gauge such factors, we propose an emulator as a statistical approximation of the human rater, and we first discuss the motivations and the rationale behind the approach.The emulator is laid down in the next section as a phenomenological model where the core affect stochastic dynamics as perceived by the rater are captured through an Ornstein–Uhlenbeck process; its parameters are then exploited to infer potential causal effects in the attribution of social traits. Following that, by resorting to a publicly available dataset, the adequacy of the model is evaluated in terms of both human raters’ emulation and machine learning predictive capabilities. We then present the results, which are followed by a general discussion concerning findings and their implications, together with advantages and potential applications of the approach.

In recent years, affective computing has emerged as a promising approach to studying user experience, replacing subjective methods that rely on participants’ self-evaluation. Affective computing uses biometrics to recognize people’s emotional states as they interact with a product. However, the cost of medical-grade biofeedback systems is prohibitive for researchers with limited budgets. An alternative solution is to use consumer-grade devices, which are more affordable. However, these devices require proprietary software to collect data, complicating data processing, synchronization, and integration. Additionally, researchers need multiple computers to control the biofeedback system, increasing equipment costs and complexity. To address these challenges, we developed a low-cost biofeedback platform using inexpensive hardware and open-source libraries. Our software can serve as a system development kit for future studies. We conducted a simple experiment with one participant to validate the platform’s effectiveness, using one baseline and two tasks that elicited distinct responses. Our low-cost biofeedback platform provides a reference architecture for researchers with limited budgets who wish to incorporate biometrics into their studies. This platform can be used to develop affective computing models in various domains, including ergonomics, human factors engineering, user experience, human behavioral studies, and human–robot interaction.

In recent years, there have been many approaches to using robots to teach computer programming. In intelligent tutoring systems and computer-aided learning, there is also some research to show that affective feedback to the student increases learning efficiency. However, a few studies on the role of incorporating an emotional personality in the robot in robot-assisted learning have found different results. To explore this issue further, we conducted a pilot study to investigate the effect of positive verbal encouragement and non-verbal emotive behaviour of the Miro-E robot during a robot-assisted programming session. The participants were tasked to program the robot’s behaviour. In the experimental group, the robot monitored the participants’ emotional state via their facial expressions, and provided affective feedback to the participants after completing each task. In the control group, the robot responded in a neutral way. The participants filled out a questionnaire before and after the programming session. The results show a positive reaction of the participants to the robot and the exercise. Though the number of participants was small, as the experiment was conducted during the pandemic, a qualitative analysis of the data was carried out. We found that the greatest affective outcome of the session was for students who had little experience or interest in programming before. We also found that the affective expressions of the robot had a negative impact on its likeability, revealing vestiges of the uncanny valley effect.

The article deals with the detection of stress using the electrodermal activity (EDA) signal measured at the wrist. We present an approach for feature extraction from EDA. The approach uses frequency spectrum analysis in multiple frequency bands. We evaluate the proposed approach using the 4 Hz EDA signal measured at the wrist in the publicly available Wearable Stress and Affect Detection (WESAD) dataset. Seven existing approaches to stress detection using EDA signals measured by wrist-worn sensors are analysed and the reported results are compared with ours. The proposed approach represents an improvement in accuracy over the other techniques studied. Moreover, we focus on time to detection (TTD) and show that our approach is able to outperform competing techniques, with fewer data points. The proposed feature extraction is computationally inexpensive, thus the presented approach is suitable for use in real-world wearable applications where both short response times and high detection performance are important. We report both binary (stress vs. no stress) as well as three-class (baseline/stress/amusement) results.

Driver monitoring systems play an important role in lower to mid-level autonomous vehicles. Our work focuses on the detection of cognitive load as a component of driver-state estimation to improve traffic safety. By inducing single and dual-task workloads of increasing intensity on 51 subjects, while continuously measuring signals from multiple modalities, based on physiological measurements such as ECG, EDA, EMG, PPG, respiration rate, skin temperature and eye tracker data, as well as behavioral measurements such as action units extracted from facial videos, performance metrics like reaction time and subjective feedback using questionnaires, we create ADABase (Autonomous Driving Cognitive Load Assessment Database) As a reference method to induce cognitive load onto subjects, we use the well-established n-back test, in addition to our novel simulator-based k-drive test, motivated by real-world semi-autonomously vehicles. We extract expert features of all measurements and find significant changes in multiple modalities. Ultimately we train and evaluate machine learning algorithms using single and multimodal inputs to distinguish cognitive load levels. We carefully evaluate model behavior and study feature importance. In summary, we introduce a novel cognitive load test, create a cognitive load database, validate changes using statistical tests, introduce novel classification and regression tasks for machine learning and train and evaluate machine learning models.

Music is capable of conveying many emotions. The level and type of emotion of the music perceived by a listener, however, is highly subjective. In this study, we present the Music Emotion Recognition with Profile information dataset (MERP). This database was collected through Amazon Mechanical Turk (MTurk) and features dynamical valence and arousal ratings of 54 selected full-length songs. The dataset contains music features, as well as user profile information of the annotators. The songs were selected from the Free Music Archive using an innovative method (a Triple Neural Network with the OpenSmile toolkit) to identify 50 songs with the most distinctive emotions. Specifically, the songs were chosen to fully cover the four quadrants of the valence-arousal space. Four additional songs were selected from the DEAM dataset to act as a benchmark in this study and filter out low quality ratings. A total of 452 participants participated in annotating the dataset, with 277 participants remaining after thoroughly cleaning the dataset. Their demographic information, listening preferences, and musical background were recorded. We offer an extensive analysis of the resulting dataset, together with a baseline emotion prediction model based on a fully connected model and an LSTM model, for our newly proposed MERP dataset.

The monitoring of emotional state is important in the prevention and management of mental health problems and is increasingly being used to support affective computing. As such, researchers are exploring various modalities from which emotion can be inferred, such as through facial images or via electroencephalography (EEG) signals. Current research commonly investigates the performance of machine-learning-based emotion recognition systems by exposing users to stimuli that are assumed to elicit a single unchanging emotional response. Moreover, in order to demonstrate better results, many models are tested in evaluation frameworks that do not reflect realistic real-world implementations. Consequently, in this paper, we explore the design of EEG-based emotion recognition systems using longer, variable stimuli using the publicly available AMIGOS dataset. Feature engineering and selection results are evaluated across four different cross-validation frameworks, including versions of leave-one-movie-out (testing with a known user, but a previously unseen movie), leave-one-person-out (testing with a known movie, but a previously unseen person), and leave-one-person-and-movie-out (testing on both a new user and new movie). Results of feature selection lead to a 13% absolute improvement over comparable previously reported studies, and demonstrate the importance of evaluation framework on the design and performance of EEG-based emotion recognition systems.

Deep learning-based emotion recognition using EEG has received increasing attention in recent years. The existing studies on emotion recognition show great variability in their employed methods including the choice of deep learning approaches and the type of input features. Although deep learning models for EEG-based emotion recognition can deliver superior accuracy, it comes at the cost of high computational complexity. Here, we propose a novel 3D convolutional neural network with a channel bottleneck module (CNN-BN) model for EEG-based emotion recognition, with the aim of accelerating the CNN computation without a significant loss in classification accuracy. To this end, we constructed a 3D spatiotemporal representation of EEG signals as the input of our proposed model. Our CNN-BN model extracts spatiotemporal EEG features, which effectively utilize the spatial and temporal information in EEG. We evaluated the performance of the CNN-BN model in the valence and arousal classification tasks. Our proposed CNN-BN model achieved an average accuracy of 99.1% and 99.5% for valence and arousal, respectively, on the DEAP dataset, while significantly reducing the number of parameters by 93.08% and FLOPs by 94.94%. The CNN-BN model with fewer parameters based on 3D EEG spatiotemporal representation outperforms the state-of-the-art models. Our proposed CNN-BN model with a better parameter efficiency has excellent potential for accelerating CNN-based emotion recognition without losing classification performance.

Data augmentation techniques have recently gained more adoption in speech processing, including speech emotion recognition. Although more data tend to be more effective, there may be a trade-off in which more data will not provide a better model. This paper reports experiments on investigating the effects of data augmentation in speech emotion recognition. The investigation aims at finding the most useful type of data augmentation and the number of data augmentations for speech emotion recognition in various conditions. The experiments are conducted on the Japanese Twitter-based emotional speech and IEMOCAP datasets. The results show that for speaker-independent data, two data augmentations with glottal source extraction and silence removal exhibited the best performance among others, even with more data augmentation techniques. For the text-independent data (including speaker and text-independent), more data augmentations tend to improve speech emotion recognition performances. The results highlight the trade-off between the number of data augmentations and the performance of speech emotion recognition showing the necessity to choose a proper data augmentation technique for a specific condition.

The study of understanding sentiment and emotion in speech is a challenging task in human multimodal language. However, in certain cases, such as telephone calls, only audio data can be obtained. In this study, we independently evaluated sentiment analysis and emotion recognition from speech using recent self-supervised learning models—specifically, universal speech representations with speaker-aware pre-training models. Three different sizes of universal models were evaluated for three sentiment tasks and an emotion task. The evaluation revealed that the best results were obtained with two classes of sentiment analysis, based on both weighted and unweighted accuracy scores (81% and 73%). This binary classification with unimodal acoustic analysis also performed competitively compared to previous methods which used multimodal fusion. The models failed to make accurate predictionsin an emotion recognition task and in sentiment analysis tasks with higher numbers of classes. The unbalanced property of the datasets may also have contributed to the performance degradations observed in the six-class emotion, three-class sentiment, and seven-class sentiment tasks.

Physical exercise has become an essential tool for treating various non-communicable diseases (also known as chronic diseases). Due to this, physical exercise allows to counter different symptoms and reduce some risk of death factors without medication. A solution to support people in doing exercises is to use artificial systems that monitor their exercise progress. While one crucial aspect is to monitor the correct physical motions for rehabilitative exercise, another essential element is to give encouraging feedback during workouts. A coaching system can track a user’s exhaustion and give motivating feedback accordingly to boost exercise adherence. For this purpose, this research investigates whether it is possible to predict the subjective exhaustion level based on non-invasive and non-wearable technology. A novel data set was recorded with the facial record as the primary predictor and individual exhaustion levels as the predicted variable. 60 participants (30 male, 30 female) took part in the data recording. 17 facial action units (AU) were extracted as predictor variables for the perceived subjective exhaustion measured using the BORG scale. Using the predictor and the target variables, several regression and classification methods were evaluated aiming to predict exhaustion. The results showed that the decision tree and support vector methods provide reasonable prediction results. The limitation of the results, depending on participants being in the training data set and subjective variables (e.g., participants smiling during the exercises) were further discussed.

In recent years, a number of new research papers have emerged on the application of neural networks in affective computing. One of the newest trends observed is the utilization of graph neural networks (GNNs) to recognize emotions. The study presented in the paper follows this trend. Within the work, GraphSleepNet (a GNN for classifying the stages of sleep) was adjusted for emotion recognition and validated for this purpose. The key assumption of the validation was to analyze its correctness for the Circumplex model to further analyze the solution for emotion recognition in the Ekman modal. The novelty of this research is not only the utilization of a GNN network with GraphSleepNet architecture for emotion recognition, but also the analysis of the potential of emotion recognition based on differential entropy features in the Ekman model with a neutral state and a special focus on continuous emotion recognition during the performance of an activity The GNN was validated against the AMIGOS dataset. The research shows how the use of various modalities influences the correctness of the recognition of basic emotions and the neutral state. Moreover, the correctness of the recognition of basic emotions is validated for two configurations of the GNN. The results show numerous interesting observations for Ekman’s model while the accuracy of the Circumplex model is similar to the baseline methods.

The emotional speech recognition method presented in this article was applied to recognize the emotions of students during online exams in distance learning due to COVID-19. The purpose of this method is to recognize emotions in spoken speech through the knowledge base of emotionally charged words, which are stored as a code book. The method analyzes human speech for the presence of emotions. To assess the quality of the method, an experiment was conducted for 420 audio recordings. The accuracy of the proposed method is 79.7% for the Kazakh language. The method can be used for different languages and consists of the following tasks: capturing a signal, detecting speech in it, recognizing speech words in a simplified transcription, determining word boundaries, comparing a simplified transcription with a code book, and constructing a hypothesis about the degree of speech emotionality. In case of the presence of emotions, there occurs complete recognition of words and definitions of emotions in speech. The advantage of this method is the possibility of its widespread use since it is not demanding on computational resources. The described method can be applied when there is a need to recognize positive and negative emotions in a crowd, in public transport, schools, universities, etc. The experiment carried out has shown the effectiveness of this method. The results obtained will make it possible in the future to develop devices that begin to record and recognize a speech signal, for example, in the case of detecting negative emotions in sounding speech and, if necessary, transmitting a message about potential threats or riots.

Every human being experiences emotions daily, e.g., joy, sadness, fear, anger. These might be revealed through speech—words are often accompanied by our emotional states when we talk. Different acoustic emotional databases are freely available for solving the Emotional Speech Recognition (ESR) task. Unfortunately, many of them were generated under non-real-world conditions, i.e., actors played emotions, and recorded emotions were under fictitious circumstances where noise is non-existent. Another weakness in the design of emotion recognition systems is the scarcity of enough patterns in the available databases, causing generalization problems and leading to overfitting. This paper examines how different recording environmental elements impact system performance using a simple logistic regression algorithm. Specifically, we conducted experiments simulating different scenarios, using different levels of Gaussian white noise, real-world noise, and reverberation. The results from this research show a performance deterioration in all scenarios, increasing the error probability from 25.57% to 79.13% in the worst case. Additionally, a virtual enlargement method and a robust multi-scenario speech-based emotion recognition system are proposed. Our system’s average error probability of 34.57% is comparable to the best-case scenario with 31.55%. The findings support the prediction that simulated emotional speech databases do not offer sufficient closeness to real scenarios.

Extensive possibilities of applications have rendered emotion recognition ineluctable and challenging in the fields of computer science as well as in human-machine interaction and affective computing. Fields that, in turn, are increasingly requiring real-time applications or interactions in everyday life scenarios. However, while extremely desirable, an accurate and automated emotion classification approach remains a challenging issue. To this end, this study presents an automated emotion recognition model based on easily accessible physiological signals and deep learning (DL) approaches. As a DL algorithm, a Feedforward Neural Network was employed in this study. The network outcome was further compared with canonical machine learning algorithms such as random forest (RF). The developed DL model relied on the combined use of wearables and contactless technologies, such as thermal infrared imaging. Such a model is able to classify the emotional state into four classes, derived from the linear combination of valence and arousal (referring to the circumplex model of affect’s four-quadrant structure) with an overall accuracy of 70% outperforming the 66% accuracy reached by the RF model. Considering the ecological and agile nature of the technique used the proposed model could lead to innovative applications in the affective computing field.

The automatic emotion recognition domain brings new methods and technologies that might be used to enhance therapy of children with autism. The paper aims at the exploration of methods and tools used to recognize emotions in children. It presents a literature review study that was performed using a systematic approach and PRISMA methodology for reporting quantitative and qualitative results. Diverse observation channels and modalities are used in the analyzed studies, including facial expressions, prosody of speech, and physiological signals. Regarding representation models, the basic emotions are the most frequently recognized, especially happiness, fear, and sadness. Both single-channel and multichannel approaches are applied, with a preference for the first one. For multimodal recognition, early fusion was the most frequently applied. SVM and neural networks were the most popular for building classifiers. Qualitative analysis revealed important clues on participant group construction and the most common combinations of modalities and methods. All channels are reported to be prone to some disturbance, and as a result, information on a specific symptoms of emotions might be temporarily or permanently unavailable. The challenges of proper stimuli, labelling methods, and the creation of open datasets were also identified.

In this paper, we tackle the problem of predicting the affective responses of movie viewers, based on the content of the movies. Current studies on this topic focus on video representation learning and fusion techniques to combine the extracted features for predicting affect. Yet, these typically, while ignoring the correlation between multiple modality inputs, ignore the correlation between temporal inputs (i.e., sequential features). To explore these correlations, a neural network architecture—namely AttendAffectNet (AAN)—uses the self-attention mechanism for predicting the emotions of movie viewers from different input modalities. Particularly, visual, audio, and text features are considered for predicting emotions (and expressed in terms of valence and arousal). We analyze three variants of our proposed AAN: Feature AAN, Temporal AAN, and Mixed AAN. The Feature AAN applies the self-attention mechanism in an innovative way on the features extracted from the different modalities (including video, audio, and movie subtitles) of a whole movie to, thereby, capture the relationships between them. The Temporal AAN takes the time domain of the movies and the sequential dependency of affective responses into account. In the Temporal AAN, self-attention is applied on the concatenated (multimodal) feature vectors representing different subsequent movie segments. In the Mixed AAN, we combine the strong points of the Feature AAN and the Temporal AAN, by applying self-attention first on vectors of features obtained from different modalities in each movie segment and then on the feature representations of all subsequent (temporal) movie segments. We extensively trained and validated our proposed AAN on both the MediaEval 2016 dataset for the Emotional Impact of Movies Task and the extended COGNIMUSE dataset. Our experiments demonstrate that audio features play a more influential role than those extracted from video and movie subtitles when predicting the emotions of movie viewers on these datasets. The models that use all visual, audio, and text features simultaneously as their inputs performed better than those using features extracted from each modality separately. In addition, the Feature AAN outperformed other AAN variants on the above-mentioned datasets, highlighting the importance of taking different features as context to one another when fusing them. The Feature AAN also performed better than the baseline models when predicting the valence dimension.

Smart mirrors are devices that can display any kind of information and can interact with the user using touch and voice commands. Different kinds of smart mirrors exist: general purpose, medical, fashion, and other task specific ones. General purpose smart mirrors are suitable for home environments but the exiting ones offer similar, limited functionalities. In this paper, we present a general-purpose smart mirror that integrates several functionalities, standard and advanced, to support users in their everyday life. Among the advanced functionalities are the capabilities of detecting a person’s emotions, the short- and long-term monitoring and analysis of the emotions, a double authentication protocol to preserve the privacy, and the integration of Alexa Skills to extend the applications of the smart mirrors. We exploit a deep learning technique to develop most of the smart functionalities. The effectiveness of the device is demonstrated by the performances of the implemented functionalities, and the evaluation in terms of its usability with real users.

The increasingly ageing population and the tendency to live alone have led science and engineering researchers to search for health care solutions. In the COVID 19 pandemic, the elderly have been seriously affected in addition to suffering from isolation and its associated and psychological consequences. This paper provides an overview of the RobWell (Robotic-based Well-Being Monitoring and Coaching System for the Elderly in their Daily Activities) system. It is a system focused on the field of artificial intelligence for mood prediction and coaching. This paper presents a general overview of the initially proposed system as well as the preliminary results related to the home automation subsystem, autonomous robot navigation and mood estimation through machine learning prior to the final system integration, which will be discussed in future works. The main goal is to improve their mental well-being during their daily household activities. The system is composed of ambient intelligence with intelligent sensors, actuators and a robotic platform that interacts with the user. A test smart home system was set up in which the sensors, actuators and robotic platform were integrated and tested. For artificial intelligence applied to mood prediction, we used machine learning to classify several physiological signals into different moods. In robotics, it was concluded that the ROS autonomous navigation stack and its autodocking algorithm were not reliable enough for this task, while the robot’s autonomy was sufficient. Semantic navigation, artificial intelligence and computer vision alternatives are being sought.

An intriguing challenge in the human–robot interaction field is the prospect of endowing robots with emotional intelligence to make the interaction more genuine, intuitive, and natural. A crucial aspect in achieving this goal is the robot’s capability to infer and interpret human emotions. Thanks to its design and open programming platform, the NAO humanoid robot is one of the most widely used agents for human interaction. As with person-to-person communication, facial expressions are the privileged channel for recognizing the interlocutor’s emotional expressions. Although NAO is equipped with a facial expression recognition module, specific use cases may require additional features and affective computing capabilities that are not currently available. This study proposes a highly accurate convolutional-neural-network-based facial expression recognition model that is able to further enhance the NAO robot’ awareness of human facial expressions and provide the robot with an interlocutor’s arousal level detection capability. Indeed, the model tested during human–robot interactions was 91% and 90% accurate in recognizing happy and sad facial expressions, respectively; 75% accurate in recognizing surprised and scared expressions; and less accurate in recognizing neutral and angry expressions. Finally, the model was successfully integrated into the NAO SDK, thus allowing for high-performing facial expression classification with an inference time of 0.34 ± 0.04 s.

During social interaction, humans recognize others’ emotions via individual features and interpersonal features. However, most previous automatic emotion recognition techniques only used individual features—they have not tested the importance of interpersonal features. In the present study, we asked whether interpersonal features, especially time-lagged synchronization features, are beneficial to the performance of automatic emotion recognition techniques. We explored this question in the main experiment (speaker-dependent emotion recognition) and supplementary experiment (speaker-independent emotion recognition) by building an individual framework and interpersonal framework in visual, audio, and cross-modality, respectively. Our main experiment results showed that the interpersonal framework outperformed the individual framework in every modality. Our supplementary experiment showed—even for unknown communication pairs—that the interpersonal framework led to a better performance. Therefore, we concluded that interpersonal features are useful to boost the performance of automatic emotion recognition tasks. We hope to raise attention to interpersonal features in this study.

In this research, we develop an affective computing method based on machine learning for emotion recognition using a wireless protocol and a wearable electroencephalography (EEG) custom-designed device. The system collects EEG signals using an eight-electrode placement on the scalp; two of these electrodes were placed in the frontal lobe, and the other six electrodes were placed in the temporal lobe. We performed experiments on eight subjects while they watched emotive videos. Six entropy measures were employed for extracting suitable features from the EEG signals. Next, we evaluated our proposed models using three popular classifiers: a support vector machine (SVM), multi-layer perceptron (MLP), and one-dimensional convolutional neural network (1D-CNN) for emotion classification; both subject-dependent and subject-independent strategies were used. Our experiment results showed that the highest average accuracies achieved in the subject-dependent and subject-independent cases were 85.81% and 78.52%, respectively; these accuracies were achieved using a combination of the sample entropy measure and 1D-CNN. Moreover, our study investigates the T8 position (above the right ear) in the temporal lobe as the most critical channel among the proposed measurement positions for emotion classification through electrode selection. Our results prove the feasibility and efficiency of our proposed EEG-based affective computing method for emotion recognition in real-world applications.

Affective computing is a field of study that integrates human affects and emotions with artificial intelligence into systems or devices. A system or device with affective computing is beneficial for the mental health and wellbeing of individuals that are stressed, anguished, or depressed. Emotion recognition systems are an important technology that enables affective computing. Currently, there are a lot of ways to build an emotion recognition system using various techniques and algorithms. This review paper focuses on emotion recognition research that adopted electrocardiograms (ECGs) as a unimodal approach as well as part of a multimodal approach for emotion recognition systems. Critical observations of data collection, pre-processing, feature extraction, feature selection and dimensionality reduction, classification, and validation are conducted. This paper also highlights the architectures with accuracy of above 90%. The available ECG-inclusive affective databases are also reviewed, and a popularity analysis is presented. Additionally, the benefit of emotion recognition systems towards healthcare systems is also reviewed here. Based on the literature reviewed, a thorough discussion on the subject matter and future works is suggested and concluded. The findings presented here are beneficial for prospective researchers to look into the summary of previous works conducted in the field of ECG-based emotion recognition systems, and for identifying gaps in the area, as well as in developing and designing future applications of emotion recognition systems, especially in improving healthcare.

Research on affective communication for socially assistive robots has been conducted to enable physical robots to perceive, express, and respond emotionally. However, the use of affective computing in social robots has been limited, especially when social robots are designed for children, and especially those with autism spectrum disorder (ASD). Social robots are based on cognitive-affective models, which allow them to communicate with people following social behaviors and rules. However, interactions between a child and a robot may change or be different compared to those with an adult or when the child has an emotional deficit. In this study, we systematically reviewed studies related to computational models of emotions for children with ASD. We used the Scopus, WoS, Springer, and IEEE-Xplore databases to answer different research questions related to the definition, interaction, and design of computational models supported by theoretical psychology approaches from 1997 to 2021. Our review found 46 articles; not all the studies considered children or those with ASD.

Data reusability is an important feature of current research, just in every field of science. Modern research in Affective Computing, often rely on datasets containing experiments-originated data such as biosignals, video clips, or images. Moreover, conducting experiments with a vast number of participants to build datasets for Affective Computing research is time-consuming and expensive. Therefore, it is extremely important to provide solutions allowing one to (re)use data from a variety of sources, which usually demands data integration. This paper presents the Graph Representation Integrating Signals for Emotion Recognition and Analysis (GRISERA) framework, which provides a persistent model for storing integrated signals and methods for its creation. To the best of our knowledge, this is the first approach in Affective Computing field that addresses the problem of integrating data from multiple experiments, storing it in a consistent way, and providing query patterns for data retrieval. The proposed framework is based on the standardized graph model, which is known to be highly suitable for signal processing purposes. The validation proved that data from the well-known AMIGOS dataset can be stored in the GRISERA framework and later retrieved for training deep learning models. Furthermore, the second case study proved that it is possible to integrate signals from multiple sources (AMIGOS, ASCERTAIN, and DEAP) into GRISERA and retrieve them for further statistical analysis.

As COVID-19 solidifies its presence in everyday life, the interest in mental health is growing, resulting in the necessity of sentiment analysis. A smart mirror is suitable for encouraging mental comfort due to its approachability and scalability as an in-home AI device. From the aspect of natural language processing (NLP), sentiment analysis for Korean lacks an emotion dataset regarding everyday conversation. Its significant differences from English in terms of language structure make implementation challenging. The proposed smart mirror LUX provides Korean text sentiment analysis with the deep learning model, which examines GRU, LSTM, CNN, Bi-LSTM, and Bi-GRU networks. There are four emotional labels: anger, sadness, neutral, and happiness. For each emotion, there are three possible interactive responses: reciting wise sayings, playing music, and sympathizing. The implemented smart mirror also includes more-typical functions, such as a wake-up prompt, a weather reporting function, a calendar, a news reporting function, and a clock.

Previous research has proven the strong influence of emotions on student engagement and motivation. Therefore, emotion recognition is becoming very relevant in educational scenarios, but there is no standard method for predicting students’ affects. However, physiological signals have been widely used in educational contexts. Some physiological signals have shown a high accuracy in detecting emotions because they reflect spontaneous affect-related information, which is fresh and does not require additional control or interpretation. Most proposed works use measuring equipment for which applicability in real-world scenarios is limited because of its high cost and intrusiveness. To tackle this problem, in this work, we analyse the feasibility of developing low-cost and nonintrusive devices to obtain a high detection accuracy from easy-to-capture signals. By using both inter-subject and intra-subject models, we present an experimental study that aims to explore the potential application of Hidden Markov Models (HMM) to predict the concentration state from 4 commonly used physiological signals, namely heart rate, breath rate, skin conductance and skin temperature. We also study the effect of combining these four signals and analyse their potential use in an educational context in terms of intrusiveness, cost and accuracy. The results show that a high accuracy can be achieved with three of the signals when using HMM-based intra-subject models. However, inter-subject models, which are meant to obtain subject-independent approaches for affect detection, fail at the same task.

Autism Spectrum Disorder (ASD) impacts 1 in 54 children in the US. Two-thirds of children with ASD display problem behavior. If a caregiver can predict that a child is likely to engage in problem behavior, they may be able to take action to minimize that risk. Although experts in Applied Behavior Analysis can offer caregivers recognition and remediation strategies, there are limitations to the extent to which human prediction of problem behavior is possible without the assistance of technology. In this paper, we propose a machine learning-based predictive framework, PreMAC, that uses multimodal signals from precursors of problem behaviors to alert caregivers of impending problem behavior for children with ASD. A multimodal data capture platform, M2P3, was designed to collect multimodal training data for PreMAC. The development of PreMAC integrated a rapid functional analysis, the interview-informed synthesized contingency analysis (IISCA), for collection of training data. A feasibility study with seven 4 to 15-year-old children with ASD was conducted to investigate the tolerability and feasibility of the M2P3 platform and the accuracy of PreMAC. Results indicate that the M2P3 platform was well tolerated by the children and PreMAC could predict precursors of problem behaviors with high prediction accuracies.

Clinical studies have demonstrated that spontaneous and posed smiles have spatiotemporal differences in facial muscle movements, such as laterally asymmetric movements, which use different facial muscles. In this study, a model was developed in which video classification of the two types of smile was performed using a 3D convolutional neural network (CNN) applying a Siamese network, and using a neutral expression as reference input. The proposed model makes the following contributions. First, the developed model solves the problem caused by the differences in appearance between individuals, because it learns the spatiotemporal differences between the neutral expression of an individual and spontaneous and posed smiles. Second, using a neutral expression as an anchor improves the model accuracy, when compared to that of the conventional method using genuine and imposter pairs. Third, by using a neutral expression as an anchor image, it is possible to develop a fully automated classification system for spontaneous and posed smiles. In addition, visualizations were designed for the Siamese architecture-based 3D CNN to analyze the accuracy improvement, and to compare the proposed and conventional methods through feature analysis, using principal component analysis (PCA).

In this article, we propose using personality assessment as a way to adapt affective intelligent systems. This psychologically-grounded mechanism will divide users into groups that differ in their reactions to affective stimuli for which the behaviour of the system can be adjusted. In order to verify the hypotheses, we conducted an experiment on 206 people, which consisted of two proof-of-concept demonstrations: a “classical” stimuli presentation part, and affective games that provide a rich and controllable environment for complex emotional stimuli. Several significant links between personality traits and the psychophysiological signals (electrocardiogram (ECG), galvanic skin response (GSR)), which were gathered while using the BITalino (r)evolution kit platform, as well as between personality traits and reactions to complex stimulus environment, are promising results that indicate the potential of the proposed adaptation mechanism.

In this work an affective computing approach is used to study the human-robot interaction using a social robot to validate facial expressions in the wild. Our global goal is to evaluate that a social robot can be used to interact in a convincing manner with human users to recognize their potential emotions through facial expressions, contextual cues and bio-signals. In particular, this work is focused on analyzing facial expression. A social robot is used to validate a pre-trained convolutional neural network (CNN) which recognizes facial expressions. Facial expression recognition plays an important role in recognizing and understanding human emotion by robots. Robots equipped with expression recognition capabilities can also be a useful tool to get feedback from the users. The designed experiment allows evaluating a trained neural network in facial expressions using a social robot in a real environment. In this paper a comparison between the CNN accuracy and human experts is performed, in addition to analyze the interaction, attention and difficulty to perform a particular expression by 29 non-expert users. In the experiment, the robot leads the users to perform different facial expressions in motivating and entertaining way. At the end of the experiment, the users are quizzed about their experience with the robot. Finally, a set of experts and the CNN classify the expressions. The obtained results allow affirming that the use of social robot is an adequate interaction paradigm for the evaluation on facial expression.

In recent years, emotion recognition algorithms have achieved high efficiency, allowing the development of various affective and affect-aware applications. This advancement has taken place mainly in the environment of personal computers offering the appropriate hardware and sufficient power to process complex data from video, audio, and other channels. However, the increase in computing and communication capabilities of smartphones, the variety of their built-in sensors, as well as the availability of cloud computing services have made them an environment in which the task of recognising emotions can be performed at least as effectively. This is possible and particularly important due to the fact that smartphones and other mobile devices have become the main computer devices used by most people. This article provides a systematic overview of publications from the last 10 years related to emotion recognition methods using smartphone sensors. The characteristics of the most important sensors in this respect are presented, and the methods applied to extract informative features on the basis of data read from these input channels. Then, various machine learning approaches implemented to recognise emotional states are described.

A novel interactive system for calligraphy called mind calligraphy that reflects the writer’s emotions in real time by affective computing and visualization techniques is proposed. Differently from traditional calligraphy, which emphasizes artistic expression, the system is designed to visualize the writer’s mental-state changes during writing using audio-visual tools. The writer’s mental state is measured with a brain wave machine to yield attention and meditation signals, which are classified next into the four types of emotion, namely, focusing, relaxation, calmness, and anxiety. These emotion types then are represented both by animations and color palettes for by-standing observers to appreciate. Based on conclusions drawn from data collected from on-site observations, surveys via Likert-scale questionnaires, and semi-structured interviews, the proposed system was improved gradually. The participating writers’ cognitive, emotional, and behavioral engagements in the system were recorded and analyzed to obtain the following findings: (1) the interactions with the system raise the writer’s interest in calligraphy; (2) the proposed system reveals the writer’s emotions during the writing process in real time via animations of mixtures of fish swimming and sounds of raindrops, insects, and thunder; (3) the dynamic visualization of the writer’s emotion through animations and color-palette displays makes the writer understand better the connection of calligraphy and personal emotions; (4) the real-time audio-visual feedback increases the writer’s willingness to continue in calligraphy; and (5) the engagement of the writer in the system with interactions of diversified forms provides the writer with a new experience of calligraphy.

Emotions play a critical role in our daily lives, so the understanding and recognition of emotional responses is crucial for human research. Affective computing research has mostly used non-immersive two-dimensional (2D) images or videos to elicit emotional states. However, immersive virtual reality, which allows researchers to simulate environments in controlled laboratory conditions with high levels of sense of presence and interactivity, is becoming more popular in emotion research. Moreover, its synergy with implicit measurements and machine-learning techniques has the potential to impact transversely in many research areas, opening new opportunities for the scientific community. This paper presents a systematic review of the emotion recognition research undertaken with physiological and behavioural measures using head-mounted displays as elicitation devices. The results highlight the evolution of the field, give a clear perspective using aggregated analysis, reveal the current open issues and provide guidelines for future research.

Owing to the increased public interest in passive brain–computer interface (pBCI) applications, many wearable devices for capturing electroencephalogram (EEG) signals in daily life have recently been released on the market. However, there exists no well-established criterion to determine the electrode configuration for such devices. Herein, an overall procedure is proposed to determine the optimal electrode configurations of wearable EEG devices that yield the optimal performance for intended pBCI applications. We utilized two EEG datasets recorded in different experiments designed to modulate emotional or attentional states. Emotion-specialized EEG headsets were designed to maximize the accuracy of classification of different emotional states using the emotion-associated EEG dataset, and attention-specialized EEG headsets were designed to maximize the temporal correlation between the EEG index and the behavioral attention index. General purpose electrode configurations were designed to maximize the overall performance in both applications for different numbers of electrodes (2, 4, 6, and 8). The performance was then compared with that of existing wearable EEG devices. Simulations indicated that the proposed electrode configurations allowed for more accurate estimation of the users’ emotional and attentional states than the conventional electrode configurations, suggesting that wearable EEG devices should be designed according to the well-established EEG datasets associated with the target pBCI applications.

Emotional responses are associated with distinct body alterations and are crucial to foster adaptive responses, well-being, and survival. Emotion identification may improve peoples’ emotion regulation strategies and interaction with multiple life contexts. Several studies have investigated emotion classification systems, but most of them are based on the analysis of only one, a few, or isolated physiological signals. Understanding how informative the individual signals are and how their combination works would allow to develop more cost-effective, informative, and objective systems for emotion detection, processing, and interpretation. In the present work, electrocardiogram, electromyogram, and electrodermal activity were processed in order to find a physiological model of emotions. Both a unimodal and a multimodal approach were used to analyze what signal, or combination of signals, may better describe an emotional response, using a sample of 55 healthy subjects. The method was divided in: (1) signal preprocessing; (2) feature extraction; (3) classification using random forest and neural networks. Results suggest that the electrocardiogram (ECG) signal is the most effective for emotion classification. Yet, the combination of all signals provides the best emotion identification performance, with all signals providing crucial information for the system. This physiological model of emotions has important research and clinical implications, by providing valuable information about the value and weight of physiological signals for emotional classification, which can critically drive effective evaluation, monitoring and intervention, regarding emotional processing and regulation, considering multiple contexts.

The ability to detect users’ emotions for the purpose of emotion engineering is currently one of the main endeavors of machine learning in affective computing. Among the more common approaches to emotion detection are methods that rely on electroencephalography (EEG), facial image processing and speech inflections. Although eye-tracking is fast in becoming one of the most commonly used sensor modalities in affective computing, it is still a relatively new approach for emotion detection, especially when it is used exclusively. In this survey paper, we present a review on emotion recognition using eye-tracking technology, including a brief introductory background on emotion modeling, eye-tracking devices and approaches, emotion stimulation methods, the emotional-relevant features extractable from eye-tracking data, and most importantly, a categorical summary and taxonomy of the current literature which relates to emotion recognition using eye-tracking. This review concludes with a discussion on the current open research problems and prospective future research directions that will be beneficial for expanding the body of knowledge in emotion detection using eye-tracking as the primary sensor modality.

Recent studies show that the elderly population has increased considerably in European society in recent years. This fact has led the European Union and many countries to propose new policies for caring services directed to this group. The current trend is to promote the care of the elderly in their own homes, thus avoiding inverting resources on residences. With this in mind, there are now new solutions in this direction, which try to make use of the continuous advances in computer science. This paper tries to advance in this area by proposing the use of a personal assistant to help older people at home while carrying out their daily activities. The proposed personal assistant is called ME3CA, and can be described as a cognitive assistant that offers users a personalised exercise plan for their rehabilitation. The system consists of a sensorisation platform along with decision-making algorithms paired with emotion detection models. ME3CA detects the users’ emotions, which are used in the decision-making process allowing for more precise suggestions and an accurate (and unbiased) knowledge about the users’ opinion towards each exercise.

In this paper, we investigate various machine learning classifiers used in our Virtual Reality (VR) system for treating acrophobia. The system automatically estimates fear level based on multimodal sensory data and a self-reported emotion assessment. There are two modalities of expressing fear ratings: the 2-choice scale, where 0 represents relaxation and 1 stands for fear; and the 4-choice scale, with the following correspondence: 0—relaxation, 1—low fear, 2—medium fear and 3—high fear. A set of features was extracted from the sensory signals using various metrics that quantify brain (electroencephalogram—EEG) and physiological linear and non-linear dynamics (Heart Rate—HR and Galvanic Skin Response—GSR). The novelty consists in the automatic adaptation of exposure scenario according to the subject’s affective state. We acquired data from acrophobic subjects who had undergone an in vivo pre-therapy exposure session, followed by a Virtual Reality therapy and an in vivo evaluation procedure. Various machine and deep learning classifiers were implemented and tested, with and without feature selection, in both a user-dependent and user-independent fashion. The results showed a very high cross-validation accuracy on the training set and good test accuracies, ranging from 42.5% to 89.5%. The most important features of fear level classification were GSR, HR and the values of the EEG in the beta frequency range. For determining the next exposure scenario, a dominant role was played by the target fear level, a parameter computed by taking into account the patient’s estimated fear level.

We investigate how pressure-sensitive smart textiles, in the form of a headband, can detect changes in facial expressions that are indicative of emotions and cognitive activities. Specifically, we present the Expressure system that performs surface pressure mechanomyography on the forehead using an array of textile pressure sensors that is not dependent on specific placement or attachment to the skin. Our approach is evaluated in systematic psychological experiments. First, through a mimicking expression experiment with 20 participants, we demonstrate the system’s ability to detect well-defined facial expressions. We achieved accuracies of 0.824 to classify among three eyebrow movements (0.333 chance-level) and 0.381 among seven full-face expressions (0.143 chance-level). A second experiment was conducted with 20 participants to induce cognitive loads with N-back tasks. Statistical analysis has shown significant correlations between the Expressure features on a fine time granularity and the cognitive activity. The results have also shown significant correlations between the Expressure features and the N-back score. From the 10 most facially expressive participants, our approach can predict whether the N-back score is above or below the average with 0.767 accuracy.

Physiological sensors can be used to detect changes in the emotional state of users with affective computing. This has lately been applied in the educational domain, aimed to better support learners during the learning process. For this purpose, we have developed the AICARP (Ambient Intelligence Context-aware Affective Recommender Platform) infrastructure, which detects changes in the emotional state of the user and provides personalized multisensorial support to help manage the emotional state by taking advantage of ambient intelligence features. We have developed a third version of this infrastructure, AICARP.V3, which addresses several problems detected in the data acquisition stage of the second version, (i.e., intrusion of the pulse sensor, poor resolution and low signal to noise ratio in the galvanic skin response sensor and slow response time of the temperature sensor) and extends the capabilities to integrate new actuators. This improved incorporates a new acquisition platform (shield) called PhyAS (Physiological Acquisition Shield), which reduces the number of control units to only one, and supports both gathering physiological signals with better precision and delivering multisensory feedback with more flexibility, by means of new actuators that can be added/discarded on top of just that single shield. The improvements in the quality of the acquired signals allow better recognition of the emotional states. Thereof, AICARP.V3 gives a more accurate personalized emotional support to the user, based on a rule-based approach that triggers multisensorial feedback, if necessary. This represents progress in solving an open problem: develop systems that perform as effectively as a human expert in a complex task such as the recognition of emotional states.

Simulation-based training has been proven to be a highly effective pedagogical strategy. However, misalignment between the participant’s level of expertise and the difficulty of the simulation has been shown to have significant negative impact on learning outcomes. To ensure that learning outcomes are achieved, we propose a novel framework for adaptive simulation with the goal of identifying the level of expertise of the learner, and dynamically modulating the simulation complexity to match the learner’s capability. To facilitate the development of this framework, we investigate the classification of expertise using biological signals monitored through wearable sensors. Trauma simulations were developed in which electrocardiogram (ECG) and galvanic skin response (GSR) signals of both novice and expert trauma responders were collected. These signals were then utilized to classify the responders’ expertise, successive to feature extraction and selection, using a number of machine learning methods. The results show the feasibility of utilizing these bio-signals for multimodal expertise classification to be used in adaptive simulation applications.

Affect recognition is an interdisciplinary research field bringing together researchers from natural and social sciences. Affect recognition research aims to detect the affective state of a person based on observables, with the goal to, for example, provide reasoning for the person’s decision making or to support mental wellbeing (e.g., stress monitoring). Recently, beside of approaches based on audio, visual or text information, solutions relying on wearable sensors as observables, recording mainly physiological and inertial parameters, have received increasing attention. Wearable systems enable an ideal platform for long-term affect recognition applications due to their rich functionality and form factor, while providing valuable insights during everyday life through integrated sensors. However, existing literature surveys lack a comprehensive overview of state-of-the-art research in wearable-based affect recognition. Therefore, the aim of this paper is to provide a broad overview and in-depth understanding of the theoretical background, methods and best practices of wearable affect and stress recognition. Following a summary of different psychological models, we detail the influence of affective states on the human physiology and the sensors commonly employed to measure physiological changes. Then, we outline lab protocols eliciting affective states and provide guidelines for ground truth generation in field studies. We also describe the standard data processing chain and review common approaches related to the preprocessing, feature extraction and classification steps. By providing a comprehensive summary of the state-of-the-art and guidelines to various aspects, we would like to enable other researchers in the field to conduct and evaluate user studies and develop wearable systems.

We present the first study in the literature that has aimed to determine Depression Anxiety Stress Scale (DASS) levels by analyzing facial expressions using Facial Action Coding System (FACS) by means of a unique noninvasive architecture on three layers designed to offer high accuracy and fast convergence: in the first layer, Active Appearance Models (AAM) and a set of multiclass Support Vector Machines (SVM) are used for Action Unit (AU) classification; in the second layer, a matrix is built containing the AUs’ intensity levels; and in the third layer, an optimal feedforward neural network (FFNN) analyzes the matrix from the second layer in a pattern recognition task, predicting the DASS levels. We obtained 87.2% accuracy for depression, 77.9% for anxiety, and 90.2% for stress. The average prediction time was 64 s, and the architecture could be used in real time, allowing health practitioners to evaluate the evolution of DASS levels over time. The architecture could discriminate with 93% accuracy between healthy subjects and those affected by Major Depressive Disorder (MDD) or Post-traumatic Stress Disorder (PTSD), and 85% for Generalized Anxiety Disorder (GAD). For the first time in the literature, we determined a set of correlations between DASS, induced emotions, and FACS, which led to an increase in accuracy of 5%. When tested on AVEC 2014 and ANUStressDB, the method offered 5% higher accuracy, sensitivity, and specificity compared to other state-of-the-art methods.

Emotion detection based on computer vision and remote extraction of user signals commonly rely on stimuli where users have a passive role with limited possibilities for interaction or emotional involvement, e.g., images and videos. Predictive models are also trained on a group level, which potentially excludes or dilutes key individualities of users. We present a non-obtrusive, multifactorial, user-tailored emotion detection method based on remotely estimated psychophysiological signals. A neural network learns the emotional profile of a user during the interaction with calibration games, a novel game-based emotion elicitation material designed to induce emotions while accounting for particularities of individuals. We evaluate our method in two experiments ( n = 20 and n = 62 ) with mean classification accuracy of 61.6%, which is statistically significantly better than chance-level classification. Our approach and its evaluation present unique circumstances: our model is trained on one dataset (calibration games) and tested on another (evaluation game), while preserving the natural behavior of subjects and using remote acquisition of signals. Results of this study suggest our method is feasible and an initiative to move away from questionnaires and physical sensors into a non-obtrusive, remote-based solution for detecting emotions in a context involving more naturalistic user behavior and games.