Description

This model performs emotion recognition directly from ECG signals using a hybrid CNN-BiLSTM deep learning architecture trained on the DREAMER dataset. By combining convolutional feature extraction with bidirectional temporal modeling, the model is able to capture both spatial and sequential characteristics of ECG waveforms, enabling highly reliable emotion classification. Emotions are represented through the two widely accepted affective dimensions of Arousal and Valence, offering a scientifically grounded way to interpret physiological emotional responses. With a performance of 94% accuracy for Arousal and 90% accuracy for Valence, this model provides strong predictive capability suitable for academic, research, and practical applications. Unlike typical research code that requires complex preprocessing and long training times, this package is fully ready to use. It includes the pretrained model, preprocessed datasets in .csv and .npy formats, the original DREAMER dataset in MAT format, complete MATLAB and Python preprocessing scripts, and the full Jupyter notebook used for training and evaluation. This allows users to run predictions immediately or modify the pipeline for their own studies without spending weeks preparing data. This model is ideal for researchers, students, healthcare developers, wearable technology projects, and human-computer interaction applications that require accurate physiological emotion recognition. Whether you are building an academic project, validating concepts for a startup, or exploring affective computing, this model offers a complete and efficient starting point. For support or any inquiries, please contact: zhenanong@gmail.com.