Confusion Matrix

Confusion Matrix is a performance measurement for machine learning classification

100 True Positives - Patients with heart disease classified correctly
110 True Negatives - Patients without heart disease classified correctly
40 False Negatives - Patients who have heart disease but algorithm did not classify
50 False Positives - Patients who didn’t have heart disease but algorithm classified positive

Precision : TP / (TP+FP) = 100/(100+40) = 100/140 = 71.4%
Sensitivity (Recall) : TP / (TP+FN) = 100/(100+50) = 100/150 = 66.7%
Specificity : TN / (TN+FP) = 110/(110+40) = 110/150 = 73.3%
Negative Predict value : TN / (TN+FN) = 110/(110+50) = 110/160 = 68.7%
Accuracy : TP + TN / (TP+TN+FN+FP) = (100+110)/(100+110+40+50) = 210/300 = 70%
F1-Score : 2 x Recall x Precision / (Recall + Precision) = (2 x 0.67 x 0.71) / (0.71+0.67) = 0.95/1.38 = 68.8%

(F1-Score is the harmonic mean of Precision and Recall . It gives a better measure of the incorrectly classified cases than the Accuracy Metric.)

A system with high recall but low precision returns many results, but most of its predicted labels are incorrect when compared to the training labels.
A system with high precision but low recall is just the opposite, returning very few results, but most of its predicted labels are correct when compared to the training labels.
An ideal system with high precision and high recall will return many results, with all results labeled correctly.
Accuracy -> When TP and TN are more important
F1-score -> When FN and FP are crucial.
Accuracy -> Better when the class distribution is similar
F1-score -> Better when the class distrbution is imbalanced