A Novel Baseline-Aware Dual-Stream ConvNeXt–Transformer with Differential Feature Fusion for EEG-Based Emotion Modelling and Music Recommendation

This repository presents the implementation and experimental evaluation of a novel baseline-aware dual-stream ConvNeXt–Transformer architecture with differential feature fusion, proposed to improve classification performance. The work focuses on combining the strengths of ConvNeXt for spatial feature extraction and Transformer modules for modeling long-range dependencies, followed by an effective differential feature fusion strategy. The repository contains four primary Jupyter notebooks. The first notebook, Proposed Model (Train–Validation–Test comparison).ipynb, implements the proposed architecture and evaluates its performance across training, validation, and testing datasets, enabling an analysis of learning behavior and generalization. The second notebook, Proposed Model 10 Fold CV.ipynb, performs 10-fold cross-validation on the proposed model to assess robustness and stability, with performance metrics averaged across folds. The third notebook, Baseline Models Comparison (Accuracy, Precision, Recall, F1 score, Kappa).ipynb, provides a comprehensive performance comparison between the proposed model and several baseline approaches, including SVM, CNN, CNN-BiLSTM, CNN-RNN, MLP, Capsule Network, and Cascade Forest, using metrics such as accuracy, precision, recall, F1-score, and Cohen’s kappa. The fourth notebook, Baseline Models 10 Fold CV comparison.ipynb, conducts 10-fold cross-validation for all baseline models, focusing on average accuracy to ensure fair and statistically reliable comparisons.