[Eeglablist] Questions Regarding Unsupervised Clustering on the DEAP Dataset

[Mahtab Dadkhah]

Dear EEGLab Team,

I am currently working on EEG-based emotion recognition using the DEAPdataset and applying unsupervised clustering methods (DBSCAN and HDBSCAN).

After extracting EEG features, I realized that I had mistakenly included theDEAP rating labels (Valence, Arousal, Dominance, etc.) among the inputfeatures, which introduced data leakage. After removing these labels andrerunning the clustering, the results became considerably worse.

I would appreciate your advice on two questions:

1.     In an unsupervised DEAP analysis, how can I determine whetherpoor clustering performance is mainly caused by suboptimal feature selection orby inappropriate DBSCAN/HDBSCAN parameter settings?

2.     For cluster evaluation, should Valence, Arousal, and Dominancebe assessed separately, or is there a recommended way to evaluate clustersusing multiple emotional dimensions simultaneously?

Thank you very much for your time and guidance.

Kind regards,

Mahtab Dadkhah Tehrani
PhD Researcher in Biomedical Engineering (Bioelectricity)