From Signals to Emotions: A Machine LearningFramework for Robust Emotion Classification UsingMultimodal Physiological Data

Abstract Emotion can be classified based on the physiological data and this is an important problem in the affective computing and it hasmany application in healthcare, human computer interaction and in the monitoring of mental health. A systematic comparativestudy of different machine learning models for emotion classification is proposed in this paper using a physiological signaldataset. The dataset consist of the following features heart rate (bpm), heart rate variability (hrv), QT interval, QRS duration,oxygen saturation (spo2), skin temperature, and anomaly indicators. We compare random forest, support vector machine (svm),xgboost, gradient boosting, lightgbm, catboost, adaboost with the neural network that is the ensemble of the most efficient. TheResults Show that The Highest AC has the Ensemble Model The data is presented in the Table 1 below. The accuracy of modelcomparison: The accuracy of the combination model is 93.97%, which is higher than that of other single models. The studyoffers contribution to the understanding of different machine learning approaches for Emotion categorization and the role ofensemble methods in improving the predictive accuracy. Furthermore, we are to discuss the impact of physiological anomalieson the efficiency of emotion categorization and its possible application in emotion monitoring systems. These aspects are ofgreat benefit to the field of affective computing and emotional health assessment.