Skip to content
SDGs Center UDINUS

Evidence record · UDINUS · SDGs

Data-driven Clustering of Stroke EEG Patterns Using Statistical Time-domain

Rektorat UDINUS 2026 Published 2026-10-30

Evidence record · UDINUS · SDGs

Unit
Rektorat UDINUS
Period
2026
Goals
Record
463ad2c7-ba94-492b-8328-67c2657bee94
Published
2026-10-30

Universitas Dian Nuswantoro conducted a reproducible, fully data-driven activity to organize electroencephalogram time-domain feature representations through unsupervised clustering at the epoch level. The activity involved Sulistyono, Marcelinus Yosep Teguh; De Setiadi, Rosal Ignatius Moses; Soeleman, Moch Arief; Mohamed, Mohamad Afendee; Marjuni, Aris; and Sambas, Aceng as participants. The activity used motor-task EEG recordings from four participants, with five trials for each participant. The recordings were segmented into baseline, movement, and relaxation phases using a fixed two-second sliding window with 50% overlap.

Universitas Dian Nuswantoro extracted statistical features from 2,540 constructed epochs before artifact suppression. The features included mean absolute value, standard deviation, root mean square, variance, skewness, and Hjorth parameters. Standardized artifact rejection criteria were applied through amplitude thresholding, flat-line detection, and variance deviation filtering. Following this process, 2,396 cleaned epochs were retained for clustering analysis.

The activity evaluated two complementary unsupervised algorithms, K-Means and K-Medoids, using a multidimensional internal validation framework. The framework comprised the Davies-Bouldin Index, Silhouette Score, Calinski-Harabasz Index, Dunn Index, Within-Cluster Dispersion, and Cluster Quality Ratio. The results indicated that a six-cluster configuration provided the most favorable balance between compactness and separation, achieving a minimum Davies-Bouldin Index of 0.1366 and a maximum Silhouette Score of 0.7287. Agreement across algorithms and multiple validity metrics suggested stable latent statistical structures within the standardized EEG feature space.

The activity integrated epoch-level unsupervised modeling, cross-metric validation, and explicit reproducibility documentation within a unified analytical pipeline. The documentation included epoch accounting, segmentation parameters, exclusion rules, and clustering configurations to enable independent verification of the numerical validity metrics. The identified clusters represent statistical similarity groupings and do not correspond to clinical categories, disease severity levels, or diagnostic classifications. Subject to institutional and ethical constraints, processed feature matrices and clustering scripts are available upon reasonable request.

Reference https://www.scopus.com/pages/publications/105039043602?origin=resultslist