CSE 477 Spring 2014

This is the project page for CSE 477 Spring 2014.

iPhone app update

21. May 2014 Steven Guh Project Log (0)
Last Week:
This is a weekly update on the iPhone app. I create an create incorporated some UI design from Zach. I used Core Data database to store the data from the actual device. Concretely, when iPhone sync with the device, all the data will be stored in the Core Data database in the iPhone. In the future, we could synchronize the local Cora Data databases to the cloud, so that user/authorized person can look at the data online.

This Week:
I will finish the Stats tab with all the detail statistics graph, and prepared the device synchronize flow.

Data mining on 2nd prototype

13. May 2014 Steven Guh Project Log (0)
I made an iPhone app that takes data from our 2nd prototype, and applied the same algorithm. We get similar graph and accuracy as the first prototype. I will start working on the iPhoene app that incorporates the Zach's UI design this week.

Edit: add the cycling data

Data mining part 2

6. May 2014 Steven Guh Project Log (0)
I collected accelerometer data from Peter, Zachary and myself, and applied some machine learning algorithm. I extracted three features out of the 10 Hz 3-axis accelerometer data. Those features are Average Motion Intensity (AI), Variance of Motion Intensity(VI), and Normalized Signal Magnitude Area (SMA). The diagram below show all the feature extracted from the data set, the data denote with CNN are used for CNN algorithm, which the data are condensed for that algorithm. We can get 97% test set accuracy by using CNN algorithm, and 91% by using logistic regression classifier. However, CNN algorithm is easy to affected by noise. We could remove the noise before we applied CNN, so that we get a higher accuracy. Since we already have the multiple machine learning algorithm ready, I will start working on the iOS app that interface with our device.
Edit:
I just found out I forgot to take absolute the values on the SMA calculation. However, the accuracy did not change. Below is the corrected diagram.

Data mining

20. April 2014 Steven Guh Project Log (0)
We collect some data on Tuesday by tying the accelerometer on the right hip rather than the right upper thigh. The following picture shows how we collect the system.


We collected the accelerometer data in a 5 second interval for the motion of sitting to standing, standing to sitting, walking, limping, jumping. Then we processed those time-series data to statistical and physical features. Average, variance, normalized signal magnitude area, dominant frequency, energy, averaged acceleration energy. Scatter plots in 2D feature space below shown that we could use some logistic regression to classify different activity.