Social Group Optimized Machine-Learning Based Elderly Fall detection Approach Using Interdisciplinary Time-Series Features

Fall can be life-taking for elderly people. Early recognition of fall events can reduce the impact among this people. Fall detection has become an active research area now a days. Many researches have been carried out using a lot of sensors and deep learning algorithms that are costly and computationally expensive. In this work, a fall detection system with a simple accelerometer and interdisciplinary time domain features were introduced. Three publicly available datasets were used and five machine learning classifiers were finally introduced to classify fall events. Dimensionality reduction using principal component analysis was also studied in this work. Finally, machine learning classifiers were optimized using social group optimization to find out best performing hyperparameters. The proposed architecture showed great robustness and efficiency as it achieved almost perfect accuracy, sensitivity and specificity in all three datasets.