Update: 7/22/18
We've been working hard to figure out which algorithms and parameters are optimal for the smartwatch classification. While the exact details are still being worked out, we are leaning toward a random forest algorithm. The goal is to be able to collect our own test data and then run the code on that information in order to test the reliability of the algorithm on realistic data. Unfortunately, the data that it i being trained on is very neat and clean cut which is causing very high accuracy and precision. We hope that by testing it on more randomized data that we can figure out if the random forest is actually the best option for the classification.
We are also working on finishing up our poster for the OSCAR presentation on August 3rd!
We've been working hard to figure out which algorithms and parameters are optimal for the smartwatch classification. While the exact details are still being worked out, we are leaning toward a random forest algorithm. The goal is to be able to collect our own test data and then run the code on that information in order to test the reliability of the algorithm on realistic data. Unfortunately, the data that it i being trained on is very neat and clean cut which is causing very high accuracy and precision. We hope that by testing it on more randomized data that we can figure out if the random forest is actually the best option for the classification.
We are also working on finishing up our poster for the OSCAR presentation on August 3rd!
Update: 6/28/18
We've got a lot done in the last 4 weeks of working! We've decided to make use of the ExtraSensory Dataset Repository to help supplement our data with smartwatch motion. We've also narrowed down our topic to finding a predictive classifying model to help classify arm motions using smartwatch sensors. The hope is that if it is accurate enough, the model could be used in WeLi to help figure out what a student is doing and contact the student's mentor to assist in any sort of intervention that might be necessary.
We are currently still working to clean and sort the data partially using the tutorial code provided by the repository. We will be using Weka, which a a machine learning software, to help us figure out the most accurate classifier model. We are currently leaning toward decision trees and random forest algorithms, but we will also be exploring naive bayes, support vector machines, and artificial neural networks as possible models as well.
If we are able to tackle this problem quickly, we might then work on collecting our own data and checking the accuracy of our model with new smartwatch data or work on encrypting data transfer from the watch for privacy purposes.
If you'd like to see the data set, check out: http://extrasensory.ucsd.edu/
We've got a lot done in the last 4 weeks of working! We've decided to make use of the ExtraSensory Dataset Repository to help supplement our data with smartwatch motion. We've also narrowed down our topic to finding a predictive classifying model to help classify arm motions using smartwatch sensors. The hope is that if it is accurate enough, the model could be used in WeLi to help figure out what a student is doing and contact the student's mentor to assist in any sort of intervention that might be necessary.
We are currently still working to clean and sort the data partially using the tutorial code provided by the repository. We will be using Weka, which a a machine learning software, to help us figure out the most accurate classifier model. We are currently leaning toward decision trees and random forest algorithms, but we will also be exploring naive bayes, support vector machines, and artificial neural networks as possible models as well.
If we are able to tackle this problem quickly, we might then work on collecting our own data and checking the accuracy of our model with new smartwatch data or work on encrypting data transfer from the watch for privacy purposes.
If you'd like to see the data set, check out: http://extrasensory.ucsd.edu/