Classification of dementia risk in the elderly through gait analysis with machine learning algorithms

Abstract The irreversible and progressive decline of physiological functions is known as aging. Among these changes is brain aging, which leads to cognitive decline and the onset of dementia. This directly affects memory, learning, and motor skills, reducing gait efficiency. This study aimed to investigate the feasibility of identifying and classifying the risk of dementia based on the analysis of kinematic variables related to gait in older adults using machine learning algorithms. This cross-sectional observational study examined a sample of 59 individuals aged 60 ± 8 years, divided into two groups: 26 institutionalized older adults (GI) and 33 non-institutionalized older adults (GNI), all residing in Bragança, Portugal. Gait data were collected during a 10-m walk, recorded on video, and analyzed using Kinovea software. Cognitive status was assessed using the Mini-Mental State Examination (MMSE). Python™ was used for statistical analysis and to develop machine learning models to classify dementia risk based on gait variables. The results showed that the algorithmic models achieved an overall accuracy of 74.6%, with the AdaBoost algorithm performing best at 83.5%. Cross-validation revealed an overall accuracy of 72%, with the Support Vector Machine (SVM) classifier achieving the highest individual performance at 80%, correctly classifying 80% of cases across different data subsets. In conclusion, gait analysis combined with machine learning algorithms demonstrated a strong relationship between gait variables and dementia, proving to be a safe and efficient technique for dementia classification. This approach offers a low-cost and accessible early identification and intervention method, with potential applications in clinical and public health settings.