Identification of Walking Balance using Acceleration Sensors with MiniROCKET

Abstract

The risk of falling increases with age, affecting approximately one in three individuals over 65 and one in two over 80 annually . In Japan, the fall rate among older adults ranges from 8.5% to 25.3%, rising with age. Falls are a leading cause of fractures such as proximal femoral fractures and a major contributor to long-term care needs. Causes include environmental risks (e.g., slippery floors, poor lighting) and individual factors such as balance impairments, muscle weakness, and visual deficits. Balance function—maintaining the body’s center of gravity within the base of support—is closely linked to fall risk. To address this, we propose a method to classify balance states during walking using acceleration data collected from four waist-mounted sensors. A MiniROCKET-based model was trained to distinguish between three states: weight on the left foot, weight on the right foot, and normal walking. We also conducted a preliminary investigation on the optimal number and placement of sensors. This study aims to contribute to fall prevention by enabling real-time assessment of balance conditions during walking.