Generalized Linear Model Part 3

Binary Logistic Regression

Models the odds for “success” for response variable y depend on a set of exploratory variables

Confusion Matrix

Summarizes the performance of a binary classifier by comparing predicted labels to the actual labels of the data

TP or True Positive

Correctly predicted as Positive

TN or True Negative

Correctly predicted as Negative

FP or False Positive

Incorrectly predicted as Positive

FN or False Negative

Incorrectly predicted as Negative

Default Confusion Matrix

Accuracy

Measures of overall correctness of the classifier

Proportion of correctly classified instances over the total number of instances

Precision

Measures overall accuracy of positive predictions made by the classifier

Of all the instances classified as positive, how many were genuinely positive?

Sensitivity

True Positive Rate

Identify all the positive instances

Of all the positive instances in the dataset, how many did the classifier identify correctly?

Specificity

True Negative Rate

Identify all the negative instances

Of all the negative instances in the dataset, how many did the classifier correctly identify

False Positive Rate (FPR)

Fall Out

Negative instances that were incorrectly classified as positive

avoid false alarms in negative instances

False Negative Rate (FNR)

Miss Rate

The proportion of Positive instances that were classified as negative

avoid missing positive instance

Positive Predicted Value (PPV)

Accuracy of positive prediction

Same Formula with PRECISION

Negative Predicted Value (NPV)

Precision for the negative class

Accuracy of negative predictions, specifically.

Matthew Correlation Coefficient (MCC)

Least influenced by imbalanced ness
All four components of confusion Matrix

Ranges from -1 to +1

• +1 = Represent a perfect classifier

• 0 = indicates a random classifier

• -1 = Denotes a classifier that performs exactly the opposite to the desired behavior

Higher value = Better Classifier

Cohen’s Kappa Statistic

Insight on gain using a model

Designed model’s reliability in terms of agreement

Youden’s J statistic

probability of an informed decision

A value of 1 indicates that there are no false positives or false negatives; the model is Perfect

Receiver Operating Characteristic (ROC) Curve

plotting Sensitivity against FPR or Fall-Out

Area Under the ROC Curve (AUC)

probability that a model ranks a random positive example more highly