Abstract:
Mobile phones now days have different kinds of sensors such as accelerometer, gyroscope, proximity sensors, barometer etc. Nowadays most people have smart phones and almost all smart phones have these sensors. So data collected with these sensors can help to do some interesting thing like human action recognition. Now our phones has tri-axial accelerometer (X, Y and Z axis) and a tri-axial (X, Y and Z axis) gyroscope that measures linear acceleration and angular velocity on all three axis respectively. Data collected with these sensors can help in human activity recognition, in real time, and is quite a significant challenge for those attempting to find out calories burnt, tracking hours slept and so forth. Smart watches also have gyroscope and accelerometer to track human activity, calories burnt etc. These watches have other sensors like heart rate monitor to do or provide additional functionalities. This report compares the result of various classical machine learning paradigm such as SVM, Logistic Regression etc on engineered features and LSTM, GRU etc on raw data. Along with comparison, the ensemble model is also designed using best models among LSTM, GRU, SVM, Logistic Regression etc that aim to produce more better and improved results than the existing models and to be the state-of-art.
