your video is helpfull.you r saying that highly correlated features with the target.we no need to perform any transformation.will this impact the accuracy of the model.plz reply
Hello from Turkey! That's great content. I'd like to ask you what if our target (y) is skewed what to do then ? For example let's say I made a car price prediction model and variables are horsepower of car and color of car and car price is skewed in dataset. I did sqrt or log to car price and trained the model, got the mse and did model tuning everything is finished. And now I want to see a 180 hp black car's price. If I insert the values it will give me the sqrt'ed or log'd value right ? So if I do the reverse of log or sqrt will it give me the real car price ? Or should I do other operations ? Thank you...
Hello, if the target is skewed, you can apply log or sqrt transformation. Now the predicted value you get is not original price value, it is either sqrt or log of the price. To get the actual price from predicted value, you can apply the inverse function, for eg if transformation is square root, apply square or if transformation is log (base e) apply exponential function i.e np.exp(). I hope that helps, Thank you.
sir how do i find skewness of a list given as price = [14751, 16422, 15398, 9445, 12589, 11687, 10692, 8475, 11184, 9961, 12898, 11905] please do reply sir will be very helpful :-)
Can't say that, but removing outliers will definitely reduce the magnitude of skewness, but no guarantee on totally removing the skewness. Skewness and outliers are 2 different things, A normal distribution with 0 skewness can have outliers on both the extreme ends.
removing outliers by trimming is removing records which comes in outlier i.e. you are loosing the data. removing outliers with capping is good practice and you can try, it works very well.
in heat map one feature has 0.002, 2nd has 0.0017.when i am going to skew, getting negative values for 2nd one -0.07723174570350672 where in first 0.2155809290498895. is it correct. why -ve values comming
