Sensor-Based Human Activities Recognition and User’s Authentication Using Machine Learning Algorithms

Abstract

Human Activity Recognition is a significant zone of machine learning research because of its uses in many fields such as health care, flexible interfaces, and a smart world, human behavior detection is gaining a lot of interest. Activities are often more abstract than words, as they're more contextually symbolic of a human's daily life. Techniques for recognizing actions from sensorgenerated data are well-developed. However, there have been few attempts to target sensor-based behavior detection. The present-age smartphone is exceptional with a cutting-edge processor, more memory storage, a long-lasting battery, and highly effective underlying sensors. This gives a chance to open up new areas of data mining for human activities recognition. The Internet of Things is a quickly developing worldview for keen urban communities that gives a way of correspondence, recognizable proof, and detecting capacities among actually circulated gadgets. With the development of the Internet of Things (IoTs), client reliance on savvy frameworks and administrations, for example, brilliant machines, cell phones, security, and medical services applications, has been expanded. This requests secure verification instruments to save the clients' protection while collaborating with savvy gadgets. In this paper, a framework has been proposed for activities recognition and user authentication. We worked out with three publically available datasets WISDM, MobiAct, and MobiAct_V2 for human activities recognition. For user authentication, we used WISDM dataset. The various feature extraction tools have been explored in this research. The performance of random forest and support vector machine classifiers was outstanding on features extracted from the Matlab tool. The highest accuracy achieved for activities recognition on WISDM, MobiAct, and MobiAct_v2 datasets was 99.50%, 98.50%, and 98.43% respectively. For user authentication, maximum accuracy achieved by utilizing the Matlab features was 95% on WISDM dataset. In comparison to current work for human activity identification and user authentication, the findings showed that the proposed conspire performed better.