Question

We obtained a large set of data on daily weather, including date, wind gust speed, sunshine duration, rain or not, temperature, and pressure. With this data, we wish to understand which factors affect whether it will rain or not on the next day.

	A.	This scenario describes a classification problem
	B.	This scenario describes a regression problem

Suppose that we have a data with 20 potential predictors. We want to run a subset selection procedure to find a single best model. Considering computational complexity, which one of the two algorithms is preferable:

	A.	forward stepwise selection
	B.	best subset selection

Suppose that we want to compare two models M1 and M2. The AIC (Akaike information criterion) value of M1 is -1005.3. And that of M2 is -1012.6. If we make a selection purely based on AIC, which model is a better one?

	A.	M2
	B.	M1

We have a data set with 90 observations. If we use this data set to perform a 10-fold cross validation, how many observations are used for training at each iteration?

	A.	81
	B.	90
	C.	9
	D.	10

Answer 1

Ans a) We obtained a large set of data on daily weather, including date, wind gust speed, sunshine duration, rain or not, temperature, and pressure. With this data, we wish to understand which factors affect whether it will rain or not on the next day. This scenario describes a classification problem because we are predicting wheather it will rain or not the next day. (A) This scenario describes a classification problem

Ans - b) Forward stepwise selection algorithmis preferable as it is a type of stepwise regression which begins with an empty model and adds in variables one by one that gives the single best improvement to your model. (A) Forward stepwise selection

Ans - c) If we make selection purely on the basis of AIC then model M2 is better as lower value of AIC indicates better fit of the model. (M2)

Ans - d) We have a data set with 90 observations. If we use this data set to perform a 10-fold cross validation, then as we are making 10 folds than there will be 90/10 = 9 observations in each fold and we will train on 9 folds and test on 1 fold. So, there will be 81 observatons for training data at each iteration. (81)