MACHINE LEARNING APPROACH TO CLASSIFY DECLINE OF COGNITIVE AND MUSCLE FUNCTION IN OLDER WOMEN - GAIT CHARACTERISTICS BASED ON THREE SPEEDS - Research summary

Aug 3, 2026

This study evaluates how accurately machine learning models can classify cognitive (Cog) and muscle function (MF) decline in women aged 65 and older using selected gait features. A total of 154 participants completed cognitive assessment, a five-times sit-to-stand test, and gait tests at three speeds (preferred, slower, and faster). Results show a random forest model reached 91.2% accuracy with all gait features and 91.9% accuracy using three selected features. Optimized gait feature extraction supports objective Cog and MF classification and assessment.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages4

Reading Time10 min

AccessFree to download

INTRODUCTION

METHODS

Participants and group definitions

RESULTS

DISCUSSION

CONCLUSION

MACHINE LEARNING APPROACH TO CLASSIFY DECLINE OF COGNITIVE AND MUSCLE FUNCTION IN OLDER WOMEN: GAIT CHARACTERISTICS BASED ON THREE SPEEDS Bohyun Kim1,2, Changhong Youm1,2*, Hwayoung Park 1, Hyejin Choi1,2, Juseon Hwang1,2, and Minsoo Kim1,2 1 Biomechanics laboratory, Dong-A University, Busan, Republic of Korea 2 Department of Health Sciences, The Graduate School of Dong-A University, Busan, Republic of Korea This study aimed to evaluate the accuracy of machine learning models based on selected gait features caused by cognitive (Cog) and muscle function (MF) declines. A total of 154 women aged 65 or older performed cognitive assessments , ﬁve times sit-to-stand test , and gait test at three speeds (preferred, slower (SWS), and faster walking speed (FWS)) . The machine learning model accuracies revealed that the random forest (RF) model had 91.2% accuracy when using all gait features and 91.9% accuracy when using the three features (walking speed and coefficient of variation of the left double support phase at FWS and right double support phase at SWS) selected for Cog+MF+ and Cog–MF– classification. We suggest that machine learning analysis using selected gait features may help improve the objective classification and evaluation of Cog and MF in older women. KEYWORDS: Dementia, frailty, sarcopenia gait variability INTRODUCTION: Cognitive (Cog) and physical functions decline with natural aging (Zhou et al. , 2022), such as the degeneration of the neuromotor control system of the central nervous system (Kara et al. , 2020) . Previous studies on physical function associated with Cog decline in older adults have reported decreased gait speed (Peel et al. , 2019) and increased five sitto-stand (FSTS) times, which is associated with decreased muscle function (MF) (Noh et al. , 2020) . These results may be owing to decreased movement ability resulting from reduced motor planning and executive function with aging (Zang et al. , 2019) . Furthermore, impairments of the motor system, such as gait abnormalities and a low level of physical fitness, precede the onset of cognitive decline with age or during the early stages of dementia (Noh et al. , 2020; Peel et al. , 2019) . Gait is a controlled task that requires high levels of attention and integration of sensory input, cognition-related motor planning and execution, and the musculoskeletal system (Zang et al. , 2019) . Furthermore, recent studies have used artificial intelligence-based machine learning to improve Cog decline detection and classification using gait characteristics in older adults (Zhou et al. , 2022) . However, these studies measured speed on short walkways of 4-10 m, resulting in uncontrolled walking speeds with quantitative values, are vulnerable to data overfitting risks because of their high correlation with multiple variables (Zhou et al. , 2022) . Recent studies have recommended gait analysis of continuous steps using wearable sensors (Zhou et al. , 2022) and analysis methods that extract optimized gait characteristics to overcome these limitations (Anwary et al. , 2018) . While a previous study demonstrated the capability of wearable sensors to identify Cog decline and physical frailty (Razjouyan et al. , 2020), it has not been investigated whether wearable sensors can distinguish between groups with specific impairments, especially those with a simultaneous decline in Cog and MF. Thus, there is a need for research on objective evaluations using optimized gait feature extraction methods to simultaneously predict MF decline during the early stages of Cog decline. This study aimed to extract features using stepwise regression from gait features at three walking speeds and evaluate the accuracy of machine learning models for classifying declinesin Cog and MF. We hypothesized that these gait features would accurately classify differences in gait characteristics between groups with declines in Cog and MF and healthy groups.

FAQs

How were cognitive decline and muscle function decline defined in the study?

Cognitive decline used Korean Mini-Mental Status Examination (K-MMSE) with cutoffs for normal, mild, and severe impairment. Muscle function decline was assessed via five repetitions sit-to-stand time according to sarcopenia diagnostic criteria, and groups were defined by combinations of Cog+/- and MF+/- status.

What gait test conditions were used to build the machine learning models?

Participants performed gait tests at three walking speeds: preferred, slower (SWS), and faster (FWS). Gait features were extracted from walking at these speeds, and models were evaluated using either all features or selected subsets.

Which machine learning model performed best and what accuracy was achieved?

The random forest model performed best for the Cog+MF+ and Cog–MF– classification, reaching 91.2% accuracy using all gait features and 91.9% accuracy when using three selected gait features.

Aug 3, 2026

Upload by Theodore