Enhancing Fall Risk Assessment with Deep Learning: A CNN-BiLSTM Approach Using Plantar Pressure Sensors

Falls in older adults pose significant health risks, often resulting in severe injuries, loss of independence, reduced mobility, and increased healthcare costs. Early and accurate fall risk assessment is crucial for timely intervention and fall prevention. Traditional fall risk assessment methods, such as clinical balance tests, require expert evaluation and are time-consuming, making them impractical for large-scale screening. To address these limitations, this study proposes a novel deep learning-based fall risk assessment approach using pressure-sensing insoles.We collected plantar pressure data from 120 community-dwelling older adults aged 65 to 90 years and used Tinetti scores as ground truth labels for classification. To mitigate class imbalance issues, we applied Borderline SMOTE, which enhances model robustness and improves classification performance. Our proposed model integrates a Bidirectional Long Short-Term Memory (BiLSTM) network to capture temporal variations in Center of Pressure (COP) movements and a Convolutional Neural Network (CNN) for spatial analysis of plantar pressure distribution.Experimental results demonstrate that our hybrid CNN-BiLSTM model achieves 94% accuracy even when using only 8 sensors per foot. Additionally, Leave-One-Group-Out (LOGO) cross-validation effectively prevents overfitting by ensuring the model generalizes well across different individuals. These findings suggest that plantar pressure-based deep learning models provide an efficient, automated alternative to traditional fall risk assessments, significantly reducing the workload for healthcare professionals while maintaining high diagnostic accuracy.