Gait-Based User Identification for Smart Home: When Machine Learning Meets FBG Accelerometers

Advancements in fiber optic sensor technology and telecommunications have paved the way for innovative health monitoring systems. However, previous works in this area have often been limited by the lack of comprehensive datasets, hindering the development of accurate and robust solutions. This paper addresses this gap by presenting a novel dataset and approach to gait-based user identification using a set of four optical Fiber Bragg Grating (FBG) sensor-based accelerometers, integrated into smart home environments. By computing features such as entropy, mean, standard deviation, kurtosis, and skewness from raw signals, and employing three unsupervised machine learning models-K-means, DB-scan, and Gaussian Mixture Model (GMM)-we achieve high accuracy in distinguishing individuals. Our dataset results for each sensor are as follows: for Sensor 1, K-means with a Silhouette Score of 0.3038 and Davies-Bouldin Index of 1.1238; for Sensor 2, K-means with a Silhouette Score of 0.3543 and Davies-Bouldin Index of 0.9659; for Sensor 3, DB-scan with a Silhouette Score of 0.2650 and Davies-Bouldin Index of 0.6855; and for Sensor 4, K-means with a Silhouette Score of 0.3949 and Davies-Bouldin Index of 1.0438. This approach not only enhances user identification but also facilitates personalized healthcare applications and unobtrusive monitoring.