Classification of Elderly Fallers and Non-fallers Using Force Plate Parameters from Gait and Balance Tasks

Abstract

Accidental falls are a major health concern among older adults, hence identifying biomechanical parameters from gait and balance tasks that differentiate between fallers and non-fallers are crucial. Limited studies using force platforms to assess postural control have used machine learning algorithms to classify older adult fallers. The results are diverse due to variation in task routines, biomechanical parameters and classification algorithms. Therefore, research analysing the performance of different classification algorithms is warranted. The purpose of this study was to compare the classification accuracy of different classification algorithms for identifying elderly fallers using force plate parameters measured during balance and gait tasks. Participants included 58 non-fallers (age = 72.3 ± 5.7) and 41 fallers (age = 74.0 ± 12.3) who performed balance and gait tasks on a walkway with embedded force plates (Kistler Instruments, Winterthur, Switzerland). The force plate parameters included 2D ground reaction force (GRF)-time data and centre of pressure (COP) displacement/velocity data. Using this data as input, five different classification algorithms were used to build models: Naive Bayesian (NB), Artificial Neural Network (ANN), Linear Discriminant Analysis (LDA), Support Vector Machine (SVM) and k nearest neighbours (kNN). A maximum accuracy of 84.95% for classifying faller/non-faller categories was obtained using LDA classifier based on parameters from combined gait and balance tasks. Combining force plate parameters from gait and balance tasks resulted in higher classification accuracies of older adult fallers (>75%) for all the algorithms. The findings of this study suggest that high accuracy of classifying elderly fallers can be obtained using force plate parameters.