Human activities recognition using smart devices
Ng, Adlin Li Ting
Date of Issue2016
School of Electrical and Electronic Engineering
Today’s sensory world is all about privacy and minimal intrusion. The application of Machine Learning for Human Activities Recognition by collecting data through smart devices fits the requirements. The aim was to determine the best accuracy that can be achieved using support vector machines and also another potential classifier. The importance of features scaling, findings of whether the claim that a gyroscope could not differentiate between the sitting and the standing activities, and if gyroscope data could help the accelerometer data in boosting the learning process are areas of interests. A conventional way of Machine Learning is by the use of software packages such as liblinear, libsvm, et cetera. MATLAB’s Classification Learner app is a state-of-the-art for Machine Learning. It allows the use of several classifiers (decision trees, discriminant analysis, support vector machines, nearest neighbour, and ensemble) to train on a set of data for comparison and to obtain the best modelling. The highest accuracy that can be achieved using the support vector machines is 97.3% whereas the use of a bagged trees ensemble classifier can reach an accuracy as high as 98.1%. And without the use of features scaling results in very poor accuracy in classifying the data. It is evident from the findings that the gyroscope data did help in improving the overall classification accuracy by 3% and also, the claim that gyroscope could not differentiate between the sitting and the standing activities was proven as true from the experiment done in this report.
DRNTU::Engineering::Electrical and electronic engineering
Final Year Project (FYP)
Nanyang Technological University