Long Short Term Memory Networks for Stroke Activity Recognition base on Smartphone
Pandapotan Siagian, Erick Fernando · 2021
Aims this study to analyze rehabilitation exercises using sensor data embedded in smartphones is widely used to recognize human activities regularly to gain a better understanding of human behavior. However, it is rarely used to measure accelerators of patient movement in hospital wards. This paper proposes an optimized neural network-based feature extraction approach based on the Long Short Term Memory technique to recognize human activity using data from a tri-axial accelerometer. The activity to be measured is the Range of Motion for patient rehabilitation, and rehabilitation exercises are one of the most important steps for recovery. The results of the softmax regression trained on the dataset set have 1,076,105 records and 12 class attributes (class distribution). The experimental results on the SAR dataset show that our Long Short Term Memory-based approach is practical and successful, with an accuracy of 89.74%.