An IoT-based smart emotion recognition system by using internal body parameters

TL;DR

This paper presents an IoT-based system that uses body signals like heart rate and blood pressure to recognize emotions with high accuracy.

Contribution

A novel IoT-based emotion recognition system using internal body parameters and achieving high accuracy with Random Forest.

Findings

Random Forest achieved 90.56% accuracy and 93.34% F1-score in emotion recognition.

External validation using DEAP tasks showed 94% accuracy and strong precision, recall, and F1-score values.

The system demonstrates robustness and generalization through internal and external validation methods.

Abstract

Emotion recognition using physiological signals has gained significant attention in recent years due to its potential applications in mental health monitoring, human–computer interaction, and stress management. This study focuses on recognizing six emotional states neutral, happy, sad, fear, anger, and surprise using internal body parameters such as blood pressure, oxygen saturation, blood glucose, heart rate, and body temperature. Leveraging an Internet of Things (IoT) enabled framework, real-time data was collected from participants. An exhaustive experimental assessment has been performed on 11 different classification algorithms of the machine learning platform. Among the algorithms, the Random Forest algorithm performed better than all other algorithms with 90.56% accuracy and 93.34% F1-score. Moreover, the precision and recall of the proposed system are extremely high. Model Robustness and generalization performances were evaluated by conducting internal as well as external validation. On conducting internal validation through k-fold cross-checking, the accuracy increased to 93.18%, clearly validating the consistency in the performance of the model. Further, the external validation was conducted by using the conventional DEAP emotional tasks, showing a collective accuracy of about 94% along with very good max and weighted average precision, recall, and F1-score values for all classes of emotions. This clearly validates the efficacy of the chosen physiological features as well as the correctness of the devised approach. The findings indicate that physiological signals, combined with IoT and machine learning, provide an effective framework for emotion recognition. This research contributes to the development of real-time, non-invasive emotion recognition systems, with promising applications in healthcare, wearable devices, and personalized user experiences. Future work will explore the integration of additional physiological parameters and advanced deep-learning models for enhanced accuracy and scalability, and usage in advanced technology.

Figures