Logistic Regression

Question 1

What is logistic regression?

Answer 1

LR is a classification method for binary outcomes. it checks the outcome variable being “True”
P(y=1) = 1/(1+ e-(Bo+B1xi+B2x2+…….+BkXk)

Odds = B0+B1x1+B2x2+……..+Bkxk
e

log(Odds) = B0+B1x1+B2x2+……..+Bkxk
Bigger the Logit is , the bigger P(y=1)

Question 2

How to calculate baseline model of binary classification model?

Answer 2

get the most occurring value as a baseline value

Question 3

How to create logistic regression model in R?

Answer 3

glm(poorCare ~., family=binomial, data = Train)

Question 4

What is AIC ?

Answer 4

Its a measure of quality of model. Like adjusted R square. Only be used with same dataset. Lesser AIC is better.

Question 5

How to create predictions in LR?

Answer 5

predict(QualityLog, type = “response”) # Response gives the probabilities.

Question 6

Threshold value (t)

Answer 6

Often selected based on which errors are "better".
High threshold ( close to 1) : catches all positive cases but may miss some positive cases.
Low threshold(close to 0) : catches all negative cases  but may classify some negative cases as positive cases.
take 0.5 as a threshold if you  don't have any preference.

Question 7

Confusion matrix for selecting threshold

Answer 7

predicted = 0            predicted = 1
Actual = 0        True Negatives(TN)      False Positives(FP)
Actual = 1         False Negatives (FN)    True Positives(TP)

Question 8

Sensitivity ans Specificity

Answer 8

Sensitivity = TP/(TP + FN)
Specificity = TN /( TN + FP)

Question 9

How to select good threshold value?

Answer 9

Use ROC curve. ( in R you can use ROCR package)

table(train$poorcare, predictTrain > 0.5)

Question 10

Area Under the (ROC) Curve

Answer 10

Given a random positive or negative , proportion of time you guess which is correctly. Perfect prediction is 1.

Question 11

Outcome measures of Logistic regression.

Answer 11

Overall accuracy = (TN + TP)/Total
Sensitivity and Specificity
Overall error rate = (FP + FN)/N
False negative error rate = FN /(TP + FN)
False positive error rate = FP/(TN + FP)

Question 12

How to interpret coefficients in logistic regression?

Answer 12

1) If we have a coefficient c for a variable, then that means the log odds (or Logit) are increased by c for a unit increase in the variable.
2) If we have a coefficient c for a variable, then that means the odds are multiplied by e^c for a unit increase in the variable.

Question 13

How to interpret coefficients in logistic regression?

Answer 13

1) If we have a coefficient c for a variable, then that means the log odds (or Logit) are increased by c for a unit increase in the variable.
2) If we have a coefficient c for a variable, then that means the odds are multiplied by e^c for a unit increase in the variable.