Brain waves provide the most reliable insights into the mind’s reactions among the several markers of emotional states. Electroencephalography (EEG) serves as a vital tool for analyzing these emotion-related neural patterns. In this study, we developed an EEG-based emotional dataset featuring Bangladeshi participants. EEG signals were recorded from 40 university students using four electrode channels (TP9, AF7, AF8, and TP10) along with one amplified auxiliary (Aux) channel. Emotional responses corresponding to five target emotions—happiness, sadness, fear, anger, and neutral—were elicited using three modalities of stimuli: audio playback, video presentations, and face-to-face conversation. The resulting brainwave data were then compared and analyzed across the types. For emotion classification, Feedforward Neural Networks (FNN), Convolutional Neural Networks (CNN), Bidirectional LSTM (BiLSTM), and Gated Recurrent Unit (GRU) models were applied, achieving classification accuracies of 90%, 91%, 88.28% and 92.24% respectively.