Emotions are essential to human evolution and development, playing a vital role in decision-making, social interactions, and overall physiological health. Emotion recognition, an important facet of affective computing, has significant implications across a range of disciplines, including human-computer interaction (HCI), stress assessment, and fatigue detection etc. Recent research has shown that emotions can be quantified utilizing both external and internal data gathered from individuals. External data sources for emotion recognition include body gestures, vocal tones, facial expressions, and speech, while internal data encompasses physiological signals that reflect emotional responses within the nervous system. Importantly, ECG-based emotion recognition is often more accurate than methods relying on speech or facial expressions, given the stronger correlation between neural activity and its fluctuations in the ECG signals. This manuscript focuses on classifying emotions from ECG data, which indicates physiological changes associated with emotions through heart rate and heart rate variability (HRV). The ECG signals available in the WESAD dataset are classified into four emotional states such as baseline, stress, amusement, and meditation by employing machine learning (ML) and deep learning (DL) models. Among various ML techniques K-nearest neighbors (KNN) reported the highest accuracy of 98.99%. Convolutional neural network (CNN), artificial neural network (ANN), deep neural network (DNN), recurrent neural network (RNN), and long short term memory (LSTM) reported the accuracy of 94.34%, 95.10%, 91.07%, 96.12%, 98.81%, respectively. This innovative method assists in predicting individuals’ physiological states through wearable devices, thus enhancing brain functionality.