I am very new to data science. Just started practicing using R. I am a Tableau BI developer.
We can create a Linear Regression in Tableau to predict the total sales figure for the next 12 months. But we can do the same using R Studio with the help of Linear Regression algorithm. But whats the difference. Linear Regression is already available in any BI tool then why we need to go for Data Science Linear Regression.. ?
I know this is a very basic doubt but your advice will be a great help.
Linear Regression as a technique has been around for a long time (since about 1800) and is well understood & implemented in many tools (even Excel can run it!). It can be done with pen & paper.
As a consequence I would expect the result of a linear regression run in R and Tableau to be roughly the same.
The benefit of doing a regression in R (instead of say Tableau) is twofold:
R is primarily a statistical tool, and as such can give you insight about individual revenue drivers, with things like p values and comparing goodness of fit for different models.
R offers you alternatives to plain vanilla Linear Regression - for examle ARIMA (a time series tool) may give you valuable insights in context of sales result prediction. The range of tools available via R packages is immense.
Even though the vanilla linear regression, it is not simply the formula y = a1x1 + a2x2 blabla.....
There are a lot statistical inference in vanilla Linear regression and I’m assume it is not implemented in Tableau(I’m not a user of that, so if I’m wrong, just point It out).
For example, you may want a interval estimation instead of a point estimation, i.e. what might be a bad or good result to be happening with a relatively not small probability(confidence interval)? Do the regression coefficients really make sense?(Significance T-test)
Besides, what really important is, lm() in R is simply a function and it returns a object and you could use the result to build a analysis-chain to a further deep analysis
I am not sure about yopur question, but certainly, you could go for anovas or further analysis in R with a linear model object (like lm()). Yo can salso easily move into a glm, or glmm, or essentially, any kind of model you want to fit, using a pretty similar formula, incorporating random effects, correlation terms etc.... Of ourse, you ca build long complex models models in R and then dregde then if that is your wish, and get all the combinations and rank them by AIC. Never heard of the software you mention (I've been using R for ~15 years, and before other canned ones, or matlab)
cheers