Classification of wandering patterns in the elderly using machine learning and time series analysis

Dementia has emerged as a significant health concern due to global aging trends. A degenerative brain disorder, dementia leads to cognitive decline, memory loss, impaired communication skills, reduced abilities, and shifts in personality and mood. Dementia lacks a definitive cure, but accurate diagnosis and treatment can improve the quality of life for those affected. Wandering behavior is common in patients, and a link between wandering patterns and the severity of the disease has been established. This work addresses the challenge of detecting dementia-related wandering behaviors. The proposed strategy utilizes data imputation methods and feature extraction with the Discrete Wavelet Transformation applied to a recently developed and comprehensive dataset. Machine learning algorithms are used to perform the final detection, and hyperparameter optimization is also evaluated.Experiments show that performance achieves an accuracy of approximately 98% using the Random Forest classifier. Results are competitive with the state-of-the-art in time series classification, with improved efficiency. The proposed methodology can be used for the development of applications for dementia related research and care.