Confusion Matrix.md

Confusion Matrix

A matrix table used to describe the performance of a classification model (or "classifier") on a set of test data for which the true values are known.

There are four classes in Confusion Matrix:

1. True Positives - Class used to describe a state when True value and Predicted values are high [1 == 1]
2. False Positives - Class used to describe a state when True value is low and Predicted value is high [0 != 1]
3. True Negatives - Class used to describe a state when True value is high and Predicted value is low [1 != 0]
4. False Negatives - Class used to describe a state when True value and Predicted values are low [0 == 0]

Ex:

True Values          = [1, 1, 0, 1, 1, 1, 0, 0, 1, 0]
Predicted Values = [1, 0, 0, 1, 0, 1, 1, 1, 1, 0]

Confusion Matrix:

True/Pred	0	1
0	2	2
1	2	4

From the above matrix we can derive the following statistics:

1. Accuracy:

Accuracy of the Model:
Formula: (TP+TN)/total = (4+2) / 10 = 60%

2. Misclassification Rate (Error Rate):

How often the model is wrong.
Formula: (FP+FN)/total = (2+2) / 10 = 40%

3. Sensitivity (Recall):

Rate of True Positive - Rate of prediction of high when true values are high?
Formula: TP/totan(True(1)) = 4 / 6 = 66.6%

4. Specificity :

Rate of False Negative - Rate of prediction of low when true values are low
Formula: TN/total(True(0)) = 2 / 4 = 50%

5. Precision:

Correct Prediction of True Values
Formula: TP/total(Predicted(1)) = 4 / 6 = 66.6%

References:

1. Simple Guide for Confusion Matrix
2. Confusion Matrix - Wiki
3. SKLearn Implementation