Application of Asymmetric Windowing Recurrence Plots in ECG Signal Encoding for Emotion Classification

Human emotions are influenced by various external stimuli but can often be concealed through facial expressions. Therefore, detecting emotions using physiological signals becomes a more reliable alternative, as humans cannot easily manipulate their subconscious responses. This study aims to classify human emotional states based on electrocardiogram (ECG) signals using the LeNet-5 model. The main contribution of this research is the improved accuracy in classifying valence and arousal based on ECG signals obtained from two public datasets, namely AMIGOS and DREAMER. The Asymmetric Windowing Recurrence Plots (AWRP) encoding technique is applied to transform the cardiac signals into two-dimensional (2D) images, enabling more in-depth and accurate analysis. The results show classification using the AWRP encoding technique outperforms recurrence plot performance even with a small amount of data. The accuracy results using the AWRP encoding technique were compared with the recurrence plots encoding technique. The AMIGOS dataset achieved its best accuracy with AWRP32 at 74.79%, while the DREAMER dataset achieved its best accuracy with AWRP98 at 85.69%. The results revealed that the AWRP encoding technique had a better capability than recurrence plots in transforming time series data into 2D image data.