Machine Learning approach to Predict Cognitive Performance using HRV

Dementia has a long prodromal phase, which starts midlife with neuropathological changes. With dementia, cognitive domains are affected, leading to a decline in cognition ability. The cognitive domains that may be affected in dementia include executive function, learning, memory, social cognition, attention and language. Several studies have concentrated on drawing a link between heart rate variability (HRV) and cognitive domains. A link has been established between memory function and HRV. Individuals with a higher HRV are better placed to exercise control over their memories and are better at suppressing depressing memories. Moreover, several studies indicated that lower HRV was linked with reduced linguistic performance, poorer executive function and slower processing speed. Therefore, HRV can help when monitoring the declining functionality of these cognitive domains affected in dementia. The HRV monitoring technique is non-invasive and inexpensive. As such, HRV makes a promising non-invasive biomarker for cognitive impairment for dementia patients or those at risk of dementia. This research work employs machine learning models to investigate the relationship between HRV and cognitive performance in several cognitive domains. We tested whether physiological variables obtained from ultra-short-term (10s) HRV could predict the cognitive performance of different cognitive domains. The results showed that the support vector machine (SVM) classifier reached 82% accuracy in detecting cognitive performance, whilst Linear Discriminant Analysis successfully achieved 90% accuracy in categorising data into high and low performance between different tests.