Towards Ambient Assisted Living (AAL): Design of an IoT-based Elderly Activity Monitoring System

Abstract

This paper presents a design and development of an IoT-based system to real-time track elders' physical activities using accelerometer sensor data. The objective behind conducting such research is to overcome the lack of ability to monitor physical activities. Especially with the development of the socioeconomic sector, the number of elders who live in isolated areas such as elderly homes have increased rapidly. In such a case with declining cognitive abilities, the healthcare of these elderly personalities becomes vulnerable. This research project fulfilled the necessity of a system to capture the vital details about those people. The Internet of Things (IoT) and cloud-based applications have become a significant part of the Information and Technology sector. Realtime monitoring is a concept tightly coupled with IoT cloud cloud-native application for this application is an excellent example of that.Further, the requirement of a low-cost system was fulfilled by using hardware components such as NodeMCU and accelerometer sensors. The designed and developed system is composed of a cost-effective wrist-worn device capable of capturing hand movement on three different arises. Hence, the detected signals are transmitted to a master node to process and recognize the activity according to the detected signal. Another significant aspect of the project is using machine learning techniques to recognize the four different activities such as walking, sitting, sleeping, and standing. The use of supervised machine learning techniques is evaluated to overcome the barriers of real-time activity recognition. Further different supervised machine learning algorithms were used and evaluated, which were extracted from existing literature. The project was conducted while accomplishing the machine learning life cycle stages, and it has significantly benefitted from generating highly accurate final results for the overall system. Further different supervised machine learning algorithms were used and evaluated, which were extracted from existing literature. The supervised machine learning algorithm Decision Tree Classifier used for this study. Using the Decision Classifier Tree algorithm succeeded in gaining more than 80% of model accuracy. Since the research topic comes under a classification type-oriented problem, the testing process of the model has been done using the confusion matrix for the trained model.