Fall Detection Tool for Vertigo Patients Using The Artificial Neural Network Backpropagation Method With Telegram Notifications

Abstract

In this research, a fall detection tool was developed using a backpropagation artificial neural network (ANN) algorithm, especially for patients who are at risk of experiencing vertigo and syncope with telegram notifications. This research aims to design and implement a fall detection system for six different movements, namely sitting, walking, bending, falling sideways, falling backward, and bending using ANN Backpropagation as a decision-making system from previous research which used thresholds. The research stages were carried out by processing data and solving problems to determine the position of the accelerometer sensor, activities carried out during data collection, and characteristics of fall conditions. Data collection consisted of 120 training data and 60 test data. The results of the research show whether the recognition of falling movements has been tested using the backpropagation method at two iteration stages, namely at 10,000 epochs, and 17,000. The best results were obtained at 17,000 epoch iterations with an accuracy rate of 94%, while in the 10,000-epoch iteration the accuracy level obtained was 92%, whereas without using the method, the accuracy was 85%. The HTTP protocol was used for data communication between the device and the database server in 17 seconds so that in 1 minute 4 data is collected.