Assessment of Elderly In-Home Fall Risk Detection Using Multiple Radar Positioning and Machine Learning Classifiers

Fall detection is essential for elderly safety monitoring. Recently, radar based non-contact sensing have been explored extensively. However, multiple radar sensors based accurate fall detection is still challenging. In this work, we propose a methodology for multiple radars based fall detection using feature extraction and machine learning classifiers. For this, online publicly available radar (Rn= 5) database is considered. The radar signals are segmented, and Doppler time maps are obtained. Eighteen features are extracted and are applied to the classifiers, namely XGBoost (XGB), AdaBoost (AdaB), k-nearest neighbor (KNN) and logistic regression (LR). Experiments are performed, and performance is evaluated using 10-fold cross-validation. The proposed approach is able to discriminate elderly fall. Using the XGB classifier, the approach yields an average classification accuracy, F-measure and, kappa of 75.95%, 75.39%, 76.23% and, 0.51, respectively. Out of all radars, radar 2 yielded the highest classification results. Thus, the proposed approach could be extended for ubiquitous, real-time in-home monitoring and fall detection.