How to overcome Overfitting and Underfitting?

This short video explains why overfitting and underfitting happens mathmetically and give you insight how to resolve it.
all machine learning youtube videos from me,
• Machine Learning

[@tamajitguharoy6169]

You are really good.You can explain complex things very easily.

[@robind999]

Very good one,thanks,

[@Mankind5490]

Great explanation thank you very much!!

[@yoginderkumar2679]

감사합니다 마스터

[@dactylon1997]

nice video, thanks

[@user-gi1bm4fm8e]

고맙습니다 ㅋㅋ 저희 교수님보다 설명잘해주시네요

[@ravindarmadishetty736]

Theoretically, concept was explained good. Can you please make a video on one case study?

[@TheEasyoung]

ravindar madishetty I can’t guarantee when, but one day I will. Thanks for suggestions!

[@lizziechen2350]

Thanksss for your sharing! very clear and interesting examples which make it easy to understand. Could you plz upload more English Version Videos for business analytics?

[@TheEasyoung]

thanks a lot. I only upload what I am good at, business analytics it not my area though. I am really happy you liked the video.

[@qasimkhan-ge2tc]

Very informative video. Kindly suggest me a topic for research on overfitting problem. I need your help

[@kao9620]

Helpful video, but I couldn't get the part at 7:30 to 13:03

[@TheEasyoung]

I know learning one concept in machine learning requires more concepts to understand before. cross validation (k-fold), L1, L2 regularization, cost function concepts required to understand, I believe you can find these concepts from other blogs or youtube video. Thanks!

[@michellelee7585]

minsuk why don't you do more videos in English. lately they've all been in Korean . I really enjoy watching your videos but I don't understand Korean :(

[@TheEasyoung]

All videos from me always have english and korean version. Which video are you looking for? Thanks for watching!

[@wizzard5574]

I have a question: I want to build a classifying program, that classifies text into 2 categories: funny/not funny. I applied some features: remove stop words, remove punctuation, stem words, lemmatize words, ngram. I use LogisticRegression and SVM. I calculate the accuracy for each algorithm and I obtain 95%. After I remove stopwords, I obtain 94%, after punctuation, 92%, and with each feature the accuracy drops. If I use all of the features together, in the end I obtain an accuracy of 60% which is unbelievable. I used cross validation, and I obtained an accuracy of 93% in the end. So what is the conclusion? I have an underfitting problem? But with all of the algorithms? This is a bit strange. Sry for this text, but I need some answers and I don't know where to find them, nobody will answer me on quora or stackoverflow.

[@TheEasyoung]

Anca Elena M. Sounds like your model is overfitted to your small train data. How many train data do you have? The reason why cross validation score is way higher is because your test data difference is high. Also does your train and test data have same amount of funny and not funny data? If the data is unbalanced, accuracy is not a good measure. You may think about f1score for unbalanced data.