Very helpful video, i have a doubt, If I have a dataset, containing channels of marketing and sales, and they are indexed by start date of campaign, some marketing channels have zero values till a particular date, ie. no investments till that date but moving forward their were investment, so in this case how do I deal with week/date values, and are the zero values in this case affect the accuracy of the regression, if yes then how to deal with or impute that value..?
@@quest3604 Try these methods - Add more features, treat outliers, or transform features to the same scale. One of these should help you better improve the R Sq.
Hi, shouldn't the intercept be kept at 0? I mean it is possible, in the real world, that the business records 0 sales (or even just really low sales)... Probably in my case it is more realistic: I am using regression to forecast sales based on last year and last last year sales data, and when I did regression without blocking the intercept to 0 I had around 100 units as intercept, which realistically is not our base sales number... Any idea? Thank you for the content!
You are right. This is beginner video. For your use case, you can normalize all features and build the model. Other way is to log transform and perform Regression.
Hi! I sent you a LinkedIn request. I found this video super interesting. How about I share some real life data of google adwords campaign, (Let's say of last 5 years), and we create a data model on how much to spend on what campaign to get the best result?
@@KunaalNaik Hi! Please accept my linkedin request. We can schedule time and means to connect. I use google adwords tools and data studio to build my own predictions, and so far have been pretty accurate with them.
Very helpful video, i have a doubt, If I have a dataset, containing channels of marketing and sales, and they are indexed by start date of campaign, some marketing channels have zero values till a particular date, ie. no investments till that date but moving forward their were investment, so in this case how do I deal with week/date values, and are the zero values in this case affect the accuracy of the regression, if yes then how to deal with or impute that value..?
Yes, impute then with a mean/median or if you have a campaign that had values on same days then impute with that.
@@KunaalNaik no change observed in the R-squared value when values impute, was: 0.2773 is 0.2775, with such a low R square value, how to proceed?
@@quest3604 Try these methods - Add more features, treat outliers, or transform features to the same scale. One of these should help you better improve the R Sq.
Hi, shouldn't the intercept be kept at 0? I mean it is possible, in the real world, that the business records 0 sales (or even just really low sales)... Probably in my case it is more realistic: I am using regression to forecast sales based on last year and last last year sales data, and when I did regression without blocking the intercept to 0 I had around 100 units as intercept, which realistically is not our base sales number... Any idea? Thank you for the content!
You are right. This is beginner video. For your use case, you can normalize all features and build the model. Other way is to log transform and perform Regression.
You talked about the next video that helps to improve the model and yet no video towards that direction and now I'm stucked...
Hi! I sent you a LinkedIn request. I found this video super interesting. How about I share some real life data of google adwords campaign, (Let's say of last 5 years), and we create a data model on how much to spend on what campaign to get the best result?
@dharmadhyaksha we can do it! Lets schedule a call and see what we can do :)
@@KunaalNaik Hi! Please accept my linkedin request. We can schedule time and means to connect. I use google adwords tools and data studio to build my own predictions, and so far have been pretty accurate with them.