Machine Learning 3.2 - Linear Discriminant Analysis (LDA) and Quadratic Discriminant Analysis (QDA)

We will cover classification models in which we estimate the probability distributions for the classes. We can then compute the likelihood of each class for a new observation, and then assign the new observation to the class with the greatest likelihood. These maximum likelihood methods, such as the LDA and QDA methods you will see in this section, are often the best methods to use on data whose classes are well-approximated by standard probability distributions.
This material complements pp. 138-149 of An Introduction to Statistical Learning (faculty.marshall.usc.edu/garet....

[@gingerderidder8665]

This beats my MIT lecture. WIll be coming back for more!

[@Spiegeldondi]

A very good and concise explanation, even starting with the explanation of likelihood. Very well done!

[@lizzy1138]

Thanks for this! I needed to clarify these methods in particular, was reading about them in ISLR

[@neftalisalazar2352]

I enjoyed watching your video, thank you. I will watch more of your videos on machine learning videos thank you!

[@huilinchang8027]

Awesome lecture, thank you professor!

[@vi5hnupradeep]

Thankyou so much ! Cleared a lot of my doubts

[@Dhdhhhjjjssuxhe]

Good job. It is very easy to follow and understand

[@JappieYow]

Interesting and clear explanation! Thank you very much, this will help me in writing my thesis!

[@billbasener8784]

How did your thesis go?

[@ofal4535]

i was trying to read it my self but you made it so much simpler

[@billbasener8784]

Thanks! I am glad it was helpful.

[@geo123473]

Very great video! Thank you professor!! :)

[@zhengcao6529]

You are so great. Keep up please.

[@spencerantoniomarlen-starr3069]

10:48 ohhhhh, I was just going back and forth between the sections on LDA and QDA in three different textbooks (An Introduction to Statistical Learning, Applied Predictive Analytics, and Elements of Statistical Learning) for well over an hour and that multivariate normal pdf was really throwing me off big time. Mostly because of the capital sigma to the negative 1st power term, I didn't realize it was literally a capital sigma, I kept thinking it was a summation of something!

[@user-mw6vi2te3s]

Very useful information, thanks you professor!

[@billbasener8784]

I am glad its helpful! Thanks for the kind words.

[@jaafarelouakhchachi6170]

can you share these slides in the videos with me?

[@MrRynRules]

Thank you sir, well explained.

[@billbasener8784]

Thanks!

[@pol4624]

very good video, thank you professor

[@billbasener8784]

I am glad it is helpful. Thank you for the kind words!

[@kaym2332]

Hi! If the classes are assumed to be normally distributed, does that subsume that the features making up an observations are normally distributed as well?

[@billbasener8784]

Yes. If the each class has a multivariate normal distribution than each individual feature variable ihas a single variable normal distribution.

[@saunokchakrabarty8384]

How do you get the values of 0.15 and 0.02? I'm getting different values.

[@rmharp]

Agreed. I got approximately 0.18 and 0.003, respectively.

[@haitaoxu3468]

could you share the slide？