DTW-kernel, RMST-sparsified similarity graph with Markov Stability multiscale clustering of learner time-series

Research Method · All research methods
Evidence · 5 claims (5 for) · 1 study (1 associational), q2 · 0 of 1 report an effect size · 5 claims rest on one study

Description

A three-stage unsupervised framework for analysing online learner temporal behaviour. A dynamic time warping kernel computes pairwise similarity between raw time-series: "Our method uses a dynamic time warping similarity kernel to generate a sparsified similarity graph between learners, to which we apply a multiscale graph partitioning algorithm". RMST sparsification retains strong, globally relevant edges; Markov Stability scans Markov time to find robust partitions at all resolutions without pre-imposing cluster number or type, yielding a quasi-hierarchy from 10 down to 2 clusters that the analyst can choose from.

Accounts

• DTW-kernel, RMST-sparsified similarity graph with Markov Stability multiscale clustering of learner time-series: A three-stage unsupervised framework for analysing online learner temporal behaviour. A dynamic time warping kernel computes pairwise similarity between raw time-series: "Our method uses a dynamic time warping similarity kernel to generate a sparsified similarity graph between learners, to which we apply a multiscale graph partitioning algorithm". RMST sparsification retains strong, globally relevant edges; Markov Stability scans Markov time to find robust partitions at all resolutions without pre-imposing cluster number or type, yielding a quasi-hierarchy from 10 down to 2 clusters that the analyst can choose from. (Peach et al. (2019))

Claims

• Low-performing learners concentrate in the massed-learning ('Crammers') cluster, with 6 of 7 low performers in that cluster in the main cohort [+M]
• Unsupervised time-series clustering identifies low performers more accurately than SVM or decision-tree classifiers trained on two temporal statistics [+M]
• Bayes factors on Gaussian-process trajectories show each behavioural cluster is statistically distinct from the others [+M]
• The methodology replicates on a second cohort of 46 learners, with only 1 of 8 low performers in the distributed-learning cluster [+M]
• On the OULAD page-click dataset, 6 of 7 low performers fall in the massed-learning cluster, whose grade is significantly lower than the distributed cluster (paired t-test at 5%) [+M]

Related Research Methods

-

Key Sources

• Peach, R. L., Yaliraki, S. N., Lefevre, D. and Barahona, M. (2019). Data-driven unsupervised clustering of online learner behaviour. npj Science of Learning 4:14. https://doi.org/10.1038/s41539-019-0054-0

Page metadata

Field	Value
type	research-method
id	dtw-kernel-rmst-sparsified-similarity-graph-with-markov-stability-multiscale-clustering-of
title	DTW-kernel, RMST-sparsified similarity graph with Markov Stability multiscale clustering of learner time-series
description	A three-stage unsupervised framework for analysing online learner temporal behaviour.
status	review
generated	by process:settle-candidates on 2026-10-26
trust	unverified — no one has confirmed this page's content

Frontmatter is the page's machine-readable record — see the Schema & Guide.