Understanding why listeners enjoy music is difficult because the experience is subjective and changes as a piece unfolds. Electroencephalography (EEG) offers a non-invasive way to study neural responses during listening, yet earlier work has mostly treated music emotion as a categorical label or relied on black-box models. We formulate enjoyment prediction as a regression problem on the public MUSIN-G dataset, which contains 128-channel EEG from 20 listeners together with song-level enjoyment ratings. Six deep learning models were compared under a participant-wise protocol, in which 15 listeners were used for training and validation and 5 were held out for testing. DeepConvNet, CNN-LSTM and GNN-CNN reached the lowest mean absolute error (0.97 to 0.99 on the 1 to 5 scale), while the temporal convolutional network (TCN) had the highest Pearson correlation with the ratings (r = 0.21). Gradient-based and occlusion-based explanations, averaged over all held-out windows, identify the time segments within each two-second window that most influence the TCN predictions. The study provides a reproducible cross-listener benchmark for EEG-based enjoyment regression and shows how explainability analysis complements standard metrics.
Figures & tables
Figure 1: EEG preprocessing and analysis pipeline, showing average referencing, high-pass filtering, ICA-based artifact rejection, downsampling, channel interpolation, segmentation, and participant-wise data splitting
Parameter
Value / Description
Batch size
16
Initial learning rate
3×10−4
Optimizer
Adam / AdamW
LR scheduler
CosineAnnealing
Dropout rate
0.5–0.6
Maximum epochs
50
Table 1: Training hyperparameters applied consistently across all models
Metric
Direction
Description
MAE
Lower is better
Avg. absolute deviation
RMSE
Lower is better
Penalizes large errors
Pearson r
Higher is better
Trend alignment
Tol. Accuracy
Higher is better
Predictions within ±1
Table 2: Evaluation metrics and their purpose in enjoyment regression
Model
Spatial Modeling
Temporal Modeling
Complexity
Core Strength
Limitation
EEGNet
Depthwise filters
Short-range convolutions
Low
Parameter-efficient
Limited context
ShallowConvNet
Shallow convolutions
Band-power filtering
Low–Med
Interpretable features
Insufficient depth
DeepConvNet
Hierarchical features
Multi-scale blocks
High
Strong accuracy
Inter-subject variability
CNN–LSTM
CNN-based encoder
Recurrent modeling
High
Sequential dynamics
Training instability
TCN
Implicit integration
Dilated causal convs
Med–High
Best trend alignment
Slightly higher MAE
GNN–CNN
Graph connectivity
Temporal graph convs
High
Explicit relationships
Reduced generalization
Table 3: Architectural trade-offs and performance characteristics of evaluated deep learning models
Model
MAE ↓
RMSE ↓
Pearson r↑
Tol. Acc. (%) ↑
EEGNet
1.03
1.27
−0.01
46.7
ShallowConvNet
1.31
1.56
−0.06
35.0
DeepConvNet
0.99
1.24
0.20
51.7
CNN–LSTM
0.98
1.18
0.18
56.7
TCN
1.05
1.28
0.21
55.0
GNN–CNN
0.97
1.19
0.18
58.3
Table 4: Regression performance on the MUSIN-G dataset (15-train / 5-test participant-wise split)
Criterion
Best Model
Observation
Lowest MAE
GNN–CNN
Best absolute accuracy
Lowest RMSE
CNN–LSTM
Best against large errors
Highest Pearson r
TCN
Best trend alignment
Highest Tol. Acc.
GNN–CNN
Best proximity
Table 5: Best-performing model per evaluation criterion
Figure 2: Grad-CAM visualization highlighting temporally salient EEG regions for enjoyment prediction. Relevance is highest at the start of the window and decays over time
Figure 3: Input gradient saliency map illustrating sensitivity of TCN predictions to localized EEG signal variations across time.
Figure 4: Occlusion-based explainability analysis showing the contribution of individual temporal EEG windows to enjoyment prediction
Figure 5: Summary of XAI analysis. Temporal relevance is distributed across sustained EEG segments rather than isolated events, supporting a context-driven interpretation of enjoyment prediction
During music listening, cortical activity encodes both acoustic and expectation-related information. Prior work has shown that ANN representations resemble cortical representations and can serve as supervisory signals for EEG recognition. Here we show that distinguishing acoustic and expectation-related ANN representations as teacher targets improves EEG-based music identification. Models pretrained to predict either representation outperform non-pretrained baselines, and combining them yields complementary gains that exceed strong seed ensembles formed by varying random initializations. These findings show that teacher representation type shapes downstream performance and that representation learning can be guided by neural encoding. This work points toward advances in predictive music cognition and neural decoding. Our expectation representation, computed directly from raw signals without manual labels, reflects predictive structure beyond onset or pitch, enabling investigation of multilayer predictive encoding across diverse stimuli. Its scalability to large, diverse datasets further suggests potential for developing general-purpose EEG models grounded in cortical encoding principles.
Shogo Noguchi, Taketo Akama, Tai Nakamura +2
Sony Computer Science Laboratories, Inc., Tokyo, Japan
Recent advances in deep learning have enabled increasingly accurate electroencephalography (EEG)-based classification of Major Depressive Disorder (MDD), but the decision-making processes of high-capacity models remain difficult to interpret. This study investigates multiple post-hoc explainability methods applied to an InceptionTime architecture trained for EEG-based MDD detection. The analysis includes Shapley-based, gradient-based, and perturbation-based attribution approaches: DeepSHAP, Integrated Gradients, GradCAM, Occlusion, and Permutation Feature Importance. Explainability analysis was performed within a subject-level stratified 5-fold cross-validation framework using global attribution aggregation across EEG segments and subjects. The evaluated methods revealed partially convergent attribution patterns, with recurring emphasis on frontal, temporal, and posterior EEG regions, particularly in the right hemisphere. Quantitative comparison demonstrated substantial agreement between gradient- and perturbation-based approaches, while DeepSHAP produced comparatively distinct attribution distributions. At the same time, variability between explainability methods highlighted the influence of methodological assumptions on the resulting explanations. Overall, the results suggest that different post-hoc explainability approaches capture partially overlapping relevance structures in EEG-based deep learning models for depression detection. Although the observed attribution patterns are broadly consistent with several previous EEG studies of MDD, the analysis should be interpreted as exploratory rather than evidence of definitive neurophysiological biomarkers or clinical applicability. The study highlights both the usefulness and limitations of post-hoc explainability for interpreting black-box EEG classifiers in psychiatric applications.
Antonia Šarčević, Nikolina Frid
University of Zagreb Faculty of Electrical Engineering and Computing, Croatia
Electroencephalography (EEG) classification has shown significant promise in applications ranging from neurodegenerative disease detection to brain-computer interfaces (BCIs). Whilst deep learning methods have increasingly dominated the research landscape in recent years; these models remain largely opaque 'black boxes' with limited interpretability. In this paper, we revisit the utility of feature-based classification via a systematic pipeline designed to integrate per-electrode signal statistics with cross-brain connectivity metrics. The proposed "KnowEEG" pipeline extracts 783 per-electrode statistical features alongside between-electrode connectivity statistics, utilizing a balanced sampling strategy between high-dimensional per electrode features and connectivity. This feature-based pipeline achieves performance comparable to or exceeding state-of-the-art deep learning models across five diverse classification tasks: emotion detection, mental workload classification, eyes open/closed detection, abnormal EEG classification, and event detection. This performance is achieved without the need for Graphics Processing Unit (GPU) resources. Furthermore, the pipeline provides inherent explainability through feature importance rankings. We demonstrate via the eyes open/closed task that this explainability identifies discriminative biomarkers in the occipital region that are consistent with established neuroscience literature. These results suggest that despite significant advances in deep learning, structured feature engineering remains a high performance and interpretable method for EEG classification.