EEG Emotion Recognition Dataset_GU

This dataset comprises EEG recordings collected from 10 healthy participants (9 male, 1 female) aged 18–23 years from the Department of Information Technology, Gauhati University, India, with the aim of analyzing brain activity associated with visually induced emotions—specifically happiness, sadness, and fear. Visual stimuli were drawn from the Open Images Dataset v7 and Extensions, with each emotion block consisting of three images (5 seconds each), separated by relaxation intervals. EEG data were recorded using a 32-channel Emotiv Epoc Flex gel-based system at a sampling rate of 128 Hz, following the international 10–20 electrode placement standard. Reference electrodes were placed at P3 (CMS) and P4 (DRL), as per Emotiv's default setup. Participants first viewed a “Relax” screen (10 seconds) before emotional stimuli were presented in two phases using different image sets to increase variability. Verbal feedback was collected after each phase to confirm the participants' emotional experiences; only data from those whose self-reported emotions matched the intended labels were retained. Participants were asked to minimize movement, and multiple relaxation periods were incorporated to maintain a calm baseline. All participants provided written informed consent.The raw EEG signals were preprocessed through a structured pipeline to ensure artifact-free data suitable for analysis. First, a bandpass filter (0.5–45 Hz) was applied using zero-phase forward-backward FIR filtering to preserve the integrity of frequency components relevant to emotional states. Next, Savitzky–Golay smoothing (frame length 127, order 5) was used to remove slow-varying trends by subtracting the smoothed reference from the EEG signals. Finally, Independent Component Analysis (ICA) via EEGLAB’s runica function was employed to identify and remove ocular artifacts, retaining only clean components for analysis.