Elderly Standing Imbalance Detection Using Noise-Resilient Robust Mean Estimator and Deep Learning

Elderly standing imbalance is a critical public health concern, demanding robust and accurate detection techniques for improved safety and well-being. In this paper, we propose a novel method employing unsupervised learning and Denoising Autoencoder with Multi-Layer Perceptron networks, along with a custom adaptive Huber loss function and activation function, to classify standing states in elderly individuals. The existing Standing imbalance detection research includes difficulties such as addressing irregularities in pressure sensor data, largely stressing binary classification due to algorithmic efficiency considerations while dealing with heavy-tailed data. The approach utilizes open-source smart insole datasets, capturing left and right foot pressure data. The ensemble model DAE-MLP efficiently captures the temporal dynamics of the imbalance scores produced using the Noise-resilient robust mean estimator, enabling accurate and robust classification. This method adapts to varied degrees of data imbalance, resulting in more accurate learning. Through comprehensive evaluations, our method achieves an overall accuracy of 94 percentage on a test dataset with 53 instances. This approach serves as a proactive standing imbalance detection system for the elderly, enhancing safety and quality of life by identifying and addressing standing imbalance risks. Our research introduces an innovative solution, paving the way for advancements in elderly healthcare and safety, reducing the risk of falls and related injuries.