Nowadays, the average lifespan of people is increasing steadily. A multitude of elderly people have the tendency to reside in their own preferred dwellings. However, as a result of this, it might be too late to discover some elders' sudden abnormal situations. Therefore, it is a great concern for them to instantly and effectively deliver their distress signals to the outside world under consciousness, due to the fact that unanticipated accidents may happen all at once. This proposed system integrates the embedded system “Arduino Nicla Voice” in detection with MIT App Inventor in display on the user's interface and deploys Bluetooth Low Energy(BLE) as the communication channel. First, 8 different values of window overlap were tested and the best outcome was used in the upcoming validation method. The overall system performance in terms of accuracy, precision, recall, loss, and F1-score were examined with a 10-fold stratified cross validation. A Receiving Operating Characteristic(ROC) graph and Area Under the ROC Curve(AUC) based on predicted probability were exhibited to see the performance of each fold and the micro-average as well. Then, limitations were discussed so as to improve the robustness of this PERS system for future studies.
Read full abstract