Multi-Sensor Fusion for Emotion Recognition using Machine Learning

Emotional well-being is essential in mental and physical health, as emotional imbalances can cause psychological problems. Early detection of emotional changes is necessary to prevent severe mental health disorders and improve healthcare. Wearable sensors provide reliable physiological signals, such as Galvanic skin response (GSR), Heart rate (HR), Skin temperature (ST), and Respiratory rate (RR), which help to identify emotions more precisely than self-reported strategies. Conventional emotion detection models often struggle to process conflicting and rapidly changing physiological signals. Many existing studies use limited physiological features or recognize limited emotions, leading to lower precision. This research proposes a hybrid ML approach that combines Random Forest (RF) and XGBoost for improved emotion classification. The paper integrates multiple data sets that contain physiological signals and emotion labels, ensuring a comprehensive analysis. The data is pre-processed by reducing noise, predicting missing values, and balancing emotion labels to improve model training. The proposed model is assessed using precision, an F1 score, a classification report, and a confusion matrix to examine misclassifications. Experimental results demonstrate that this hybrid approach improves the accuracy of emotion detection and exceeds existing methods. This study contributes to advances in emotion recognition using wearable technology, providing a more efficient method of monitoring mental health in real-time and affective computing.