Feature Selection Techniques Easily Explained | Machine Learning 

This video is really the one that I was looking for all over the internet. The video was succinct and to the point.

Problems which I faced while working on this: Univariate Selection: independent variables should be non-negative, independent features shouldn't have object or category features. Feature importance: dependent variable shouldn't be a continuous variable. Correlation matrix: it will omit category and object features.

Thanks a lot for all your videos and your efforts. I would like to mention something regarding Covariance and Correlation Coeffiicients @3:25 that you talked about. What you mentioned about Correlation Coeff is correct that it ranges from -1 to 1. However, Covariance is measure of correlation and doesn't vary between 0 & 1 but varies from -inf to +inf and is unit dependent(ie, for ex: covariance of age vs height in cms will be drastically different with age vs height in inches is considered as both of them are in different units) unlike correlation coefficient where the values are standardized and hence it ranges from -1 to 1.

You are the best explainer of Machine Learning concepts Krish. I have greatly benefited form your intuitive and practical explanations. Thank you!

You definitely deserves more views! Thank you so much!

Best feature selection video on youtube. Thanks @Krish Naik

I loved ur video of feature selection. It is really informative. I just wanted to know when we can consider a data set to be huge and when we can consider it to be a smaller one ... could you please make a video or some analysis on how to handle those data sets and best possible approaches and industrial standards used on the same too?

You cover important topics, no the most important topics. So thanks!

You men deserve more more more and more likes, Just because of you I got my results. Thank you for sharing this with us.

sir,really you are a great person....now I can learn and understand better way this DS....... THANK YOU SIR..please help me for internship for data science....

Top class Quality Content it really help me to understand each and every topic easily

Very well explained . Thank you

15:17 I think there is a simpler way to select the columns for the X matrix, ie X=data.iloc[: , :-1] instead of X=data.iloc[; , 0:20]

Really super👌

Excellent

GREAT.... GREAT.... GREAT ...... EXPLAINATION KRISH BHAI

Bahut badhiya...

Really appreciate your knowledge sharing, my friend. Just only 1 thought, it would be a little better if you can slow down a little bit, for a foreigner like me it would take some time to understand and absorb.

Thanks a ton Krish.. u rock..

thank you for such videos . i really love your work . please keep uploading these videos related to data science field

Thank you so much Krish for this tutorial!

You are hero. Thanks for your explanation.

Good explanation sir 👍👌👏

Thank you so much! 1 video settles all the confusion!

Easy to understand. Thanks for the video.

Priceless lectures sir,It has well explanation with codes and beautiful examples

Best explanation...

Nicely explained...

I like it. but it is better if you leave the training data under description.

this vid was my aha moment. ty!

Thanks Krish

From above we can see that correlation matrix also give correlation between independent features,so it is more helpful than univariate selection. So can you say when we should use univariate selection over correlationmatrix and why ?

Your videos are the best! Thank you!!

Very helpful. Loved every second. Thank you very much.

awesome video sir, superb explanation

awesome tutorial! Thanks for this amazing work !

Great Insights, and now I have SUBSCRIBED, and can't wait to see more from you. Huge Thank You!!!

Your explanation helped a lot

Good job...krish, really appreciated your effort. Partical approach is always best way of learning

Wow Super B Explanation Sir

Just awesome..... damn good explanation Karish ! Thanks for the video

Very informative. Thank you. At 20:15 - Just wondering why you chose to apply correlation again by doing data[top_corr_features].corr() when its already available in corrmat. Just the fear of missing out something important. Code given below ##get correlations of each features in dataset corrmat = data.corr() top_corr_features = corrmat.index plt.figure(figsize=(20,20)) ##plot heat map g=sns.heatmap(data[top_corr_features].corr(),annot=True,cmap="RdYlGn")

If I am not mistaken, covariance and correlation coefficient measure the linear relationship between two independent random variables. If their values are zero, it mean that there is no linear relationship. But this does not mean there is no interaction between these two random variables. Instead of looking on correlation coefficient, shall we use another test that can check independence between these variables?

Sir , it is an awesome video....Very elegantly explained

Great Video. But I am not sure if Chi square test can be applied between a continuous and categorical variable. In the uni variate analysis chi square was performed between all variables including continuous against categorical target variable

very nice video

Excellent explanation Krish.. Tnx

Very nice

Great tutorial, very helpful.

thanks for this...

This was really helpful, but as I can see its a classification problem, I can use the chi-square test as far as I know. this test can be used to assess the strength of the relationship between categorical features and the output variable (which is categorical in this case ). But for not classifying problems if the data are numerical continuous data we have to use co relation. Am I right ? please help me with this doubt.

Great video Krish .. Keep uploading more and more

Hello Mister! Could you please also distinguish the difference between univariate selection and multiunivariate selection? Also, is there a difference between feature ranking, feature selection and feature extraction?

Awesome Elaboration

Nice Video... in correlation techniques with heat map we have to select features which are close to 1 or greater than 0.2

Watched full video

Krish, In general we apply chi^2 test for only categorically variables but here in the data we considers is consists of numeric data hw can we apply chi2 in univariate selection.... Can u plzzz help up with the query.

Thanks you so much for your fabulous explaination!!!

Thanks for your video and could you explain briefly how to score each feature or the philosophy behind the scoring? Or, some references are also preferable.

Great explanation. Thank you.

Doubt here: How Chi-square is working with non-categorical attributes here? In stats video, you said chi-squared is only applicable to finding relation between categorical attributes.

Very helpful. Thank you very much.

superb explain sir

you are doing very good job

You are awsome man...I am going to see all your videos

Krish, I have a doubt, we apply chi2 test for categorical features then why we have applied it for numerical features?

The best!!!!!!!!!

"and this will definitely probably work" is the best thing statisticians can say I guess :P one step better than 'probably approximately correct' )

Covariance lies between to - infinity to + infinity

Great

Perfect!

Can anyone tell me at 18;30 when krish used Extratreesclassifier and the score for every explanatory varible was generated, what is the math behind those "scores" (numbers)????

awesome sir.. Liked it n Learned from it :)

Please correct: 3:24 Covariance is not bound to vary between 0 and 1 , covariance can have values greater than 1 as well as less than 0.

awesome explaination tq sir

Great content..thanks a lot Kris

pls do a video about automatic feature selection with Featurewiz :)

It is perfect explanation

Amazing video😍

how do we apply these method when the dataset is mixed with categorical variable

Thank you so much!

your content is too good, it really help me alot, but what if the data is categorical? Will that still work or I have to convert the data into numerical form.

Which one should use all checking relationship with target

Awesome tutorial, thank you. One question that I have is, will these techniques be applicable if the data set has categorical or object/char type independent variables? The data set used here has all numeric variables.

Hi Sir, If I have (A,B,C,D,E) columns and E is my target column, then while training the model Should I use E also in X or I should use only (A,B,C,D) ?

How would you define a small - med -large dataset? Small being min 5K datapoints etc?

should we standardize or normalize data before doing this feature selection ??

Hi Krish, chi square we use for categorical variable but here few feature are continuous variable.

Hi Krish All of your videos are simple and excellent. I am not sure whether to ask you or not but if possible could you please upload the videos in 1080p format as well.

Curious to know if we should scale the dataset before applying feature selection method or after applying the feature selection method? Thank you!

Very nice video Thanks :)

Excellent one . Thanks for sharing. I have one query. In a dataset of 2 lakhs records where a variable 'country' is having high correlation with target and the country column contain all 300 country name in character format, what do you suggest. Can I use one hot encoding.

@krish naik How can you apply Chi Square test on continuous features?

You nailed it! Thank you.

Good explaination

Hi, Suppose while selecting the best features using correlation matrix, if I found out that the two features i.e. the two independent variables have a very high value of the coefficient of the correlation say 0.8 or 0.9, so shall I drop one of the features or continue working with both?

When there are multiple X-axis values such as x0 to x9 and other x values for an activity as from a gyroscope dataset which feature engg technique should we apply..

Thank for your video. can u make a video for explaining about metrics and how to choose a right metric for our algorithms???

Thank you so much❤

This is great! thanks. May I ask you why the results of my feature importance score differs every time I run the code. I am trying to get the top 30 and every time I run the code, it gives me a different set of features as the top30 largest feature importances.... am I doing something wrong?