Framework for Monitoring and Recognition of the Activities for Elderly People from Accelerometer Sensor Data Using Apache Spark

Abstract

Analysis of daily human activities becomes more open and prevalent with the advancement of sensors embedded in mobile devices. A number of applications such as fitness tracking, health analysis or user-adaptive systems, has been developed by tracking down detailed analysis of complex human activities. In this paper, a framework for the monitoring and recognition of the elderly people’s activities using Apache-Spark Big-data processing tool is proposed. In the proposed framework, different classification techniques such as Logistic Regression, Decision Tree and Random Forest Classifier of ML Machine Learning library of Apache Spark are used to recognize human activities. In order to evaluate the proposed framework, two commonly known KAGGLE-UCI and WISDM Smartphone accelerometer datasets are used. Performance analysis of classification based human activities recognition schemesis evaluated in terms of training time, testing time, accuracy and F1-score. Results show that Logistic Regression classification after tuning it with Cross Fold Validation has provided better performance compared to remaining classification methods.