Week#3 Tracking of Students’ Facial Expressions

Last week’s summary from an outsider’s point of view

What Have We Done Last Week?

Last week, we detected some mislabeled data from the dataset and created a new class called “Sleepy”.

What Have We Done This Week?

As we mentioned last week, we dealt with misclassified data by hand (unfortunately) and finalized the web scraping step for the “Sleepy” class.

Data Cleaning Process

So far, we have deleted nearly 1400 images that we think were misclassified from our dataset. We hope to finalize this step by next week. We hope to finalize this step by next week.

image0000427.jpg

For example, this image from the 7th class of our data was labeled as “Angry.” As we can all see, there’s no anger on his face.

Some of the photos we scraped for the “Sleepy” label are supposed to be less dirty. We will deal with them next week.

Data Generation Process

We used the same web scraping methods which we used last week for our “Sleepy” label and scraped from free stock photos for good measure.

Yoda’s wise advice

Conclusion

Finalized the “Sleepy” label. We cleaned up some mislabeled data.

As we further investigate new concepts for ourselves in data science, we go down the rabbit hole much deeper every time.

Thanks for your time.

by Gökçenaz Akyol and Arda Altunkaya