top of page

The Model

Utilizing the Wolfram technology, we implemented a classifier by transforming movement data into images. 

Screen Shot 2018-12-02 at 3.46.42 AM.png
Classifier was trained on 110 images labelled “walk” and “stair”. Tested on 130 images. Training and test samples were shuffled.
Screen Shot 2018-12-02 at 3.52.22 AM.png

To test robustness of the model: we misclassified specific datasets.

Screen Shot 2018-12-02 at 3.52.31 AM.png

Finally, we tested our model on a final dataset obtained by exploring different terrain around the Yale campus. The accuracy achieved was still 1.0

Thoughts?

The use of accelerometer data in determining accessibility was experimental. However with the convenience of collecting movement data as well as the utility of a baseline classifier like naive bayes performing unexpectedly well, the ease in expanding this product by using more varied datasets would be interesting. 

bottom of page