Error Metrics

Error Metrics- We have Classification Metrics and Regression Metrics

Choice of Metrics based on the type and implementation of the Model.

Classification Metrics

Confusion Matrix
Accuracy
Misclassification Error
Speicificity
Recall

With above, there are many other metrics available to evaluate classification model.

Regression Metrics

MSE
RMSE
MAE
MAPE

Confusion Matrix

* Always build Confusion Matrix on Test Data not on Train Data. **

It is used to describe the performance of the Model.
Row represents actual values and column represents predicted values.
It is used for binary or multi class classifier problem.
If our Target variable has two or more classes then we can develop confusion matrix based on that.
The matrix size will be n*n where n is number of classes in Target Variable. Below is the image of confusion matrix True Positive - Actually Positive(Yes) also Predicted Positive(Yes).
True Negative- Actually Negative(No), also predicted Negative(No).
False Positive- Actually Negative(No), but predicted Positive(Yes).
False Negative- Actually Positive(Yes), but predicted Negative(No).

Accuracy

Accuracy is measured through confusion matrix.
Accuracy = TP+ TN/ Total Observations (TP+TN+FP+FN)

Example

Misclassification Error

Misclassification Error is nothing but the error which is classifying a record as beloging to one class when it belongs to another class. And the error rate is below which is calculated from above confusion matrix.