Using Knowledge Tracing to Predict Students’ Performance in Cognitive Training and Math

• Cognitive training aims to improve skills such as working memory capacity and spatial ability, which have been linked to math skills. In this study, we fit Deep Knowledge Tracing with Transformers (DKTT), Dynamic Key-Value Memory Networks (DKVMN), and Knowledge Tracing Machines (KTM) to a large dataset from a cognitive training system. DKVMN achieved the highest AUC (0.739) of the algorithms. To explore connections between math skills and cognitive skills, the data was split into cognitive and math items. DKVMN’s AUC on the math items was higher (0.745) than on the cognitive (0.706). Notably, the split model AUCs did not differ from skill-level AUCs produced by a model trained on the entire dataset, suggesting that math performance did not improve DKVMN’s cognitive predictions and vice versa. 

Ämnesord

SAMHÄLLSVETENSKAP  -- Utbildningsvetenskap (hsv//swe)

SOCIAL SCIENCES  -- Educational Sciences (hsv//eng)

Nyckelord

Cognitive training

Knowledge tracing

Math learning

Students

Dynamic keys

Key values

Knowledge tracings

Math skills

Memory capacity

Memory network

Student performance

Working memory

Large dataset

Publikations- och innehållstyp

ref (ämneskategori)

kap (ämneskategori)