Human Activity Recognition Using Non-Invasive Wi-Fi RSSI and Hybrid Machine Learning

Human Activity Recognition (HAR) is a growing field of interest due to its wide range of applications in healthcare and fitness. One such application is daily exercise monitoring for patients to ensure they meet their prescribed exercise goals, patients suffering from obesity and prediabetes being an example. Invasive wearable technologies have been implemented in the past to track the activities but they carry major flaws, such as data security issues, physical invasiveness and collection of personally identifiable information, which has constituted to decline in the usage of the technology. Thus, this paper presents a different non-wearable non-invasive system which uses Wi-Fi-RSSI (Received Signal Strength Indicator) to detect activities using wall mounted sensor nodes and implements a lightweight supervised hybrid coarse-to-fine machine learning algorithm incorporating Nearest Centroid Classifier (NCC) and Support Vector Machine (SVM) to recognise the activities. The hybrid machine learning model emerged successful with an accuracy of 98.47%, over other separately trained models using a single algorithm like K Nearest Neighbour (KNN), Decision Tree algorithm etc.