Accuracy score is not an ideal metric when classes are skewed or not of equal importance. So, we need to optimize a slightly different version of that.
Precision: is the number of True Positives (TP) divided by the total number of positive classifications by the model (TP + FP). In essence, Precision defines how precisely the model is out of the total predicted positives. Precision is good when the cost of False Positives (FP) is high.
Recall: is the number of TP divided by the total number of actual positives (TP + FN). In essence, Recall tells how correctly the model is predicting the actual positives. Recall is a good measure when the cost of False Negatives is high.
F1 Score: is a balance between Precision and Recall. It takes the harmonic mean of the two and gives a balanced measure, which is better than accuracy. 
This is because the F1 score will be pulled down if any of the precision or recall value is low. For example, when Precision is 100% and Recall is 0%, the F1-score will be 0%, and not 50%.
But can we blindly go for F1 always? No.
Defining metrics completely depends on the domain and the data. More on this in tomorrow’s post!
#datascience #machinelearning