Data Science - Part IX - Support Vector Machine

For downloadable versions of these lectures, please go to the following link:

This lecture provides an overview of Support Vector Machines in a more relatable and accessible manner. We will go through some methods of calibration and diagnostics of SVM and then apply the technique to accurately detect breast cancer within a dataset.

[ramakanthrayanchi]

Excellent Video. I understood that projection into higher dimensions can make the non linear separators using the linear separators which i did not understand before watching this video .
Thanks for sharing.

[ajkelly]

Great video. As a caveat/disclaimer to the end of your video though, I wanted to clear a couple things up. You indicated that an accuracy of over 97% indicated how incredibly powerful SVMs are. However, these features were all handpicked and known to be indicative of separating benign versus malignant breast cancer anyways. I know, because I also worked with this dataset in recent weeks.
Another thing is that the statement itself would seem to indicate that simpler methods for classification may not be sufficient here. However, a simple logistic regression with L1 regularization gave me an AUC of 0.999 in this dataset on left-out samples. That, to me, indicates that simpler may be better here. Though the SVM did well, I still think it overfit to some extent.
In either case, thanks for the great content! You really make things clear and give great explanations.

[coolwzl]

Great content, Derek - thanks! There's a problem with the sound, though - background noise.

[pengfeijia]

Hey Derek, thanks for making the video ! Just a quick question, how to run the tuning function to identify the best parameter to use? Is there another package in R that we can employ? Or just manually pick one? Thanks in advance!

[emmygreat]

Great delivery but the audio is not great. Meanwhile, the breastcancer file is not attached, can you please share.