In this blog we are going to see about Model Evaluation of Regression Model and Classification Model in crisp and clear manner. This is one of the important topic for Azure AI-900 Certification Exam. So kindly have this blog to be a thumb rule of the Exam preparation. More over you will be learning some of the Metrix as well in this topic.

This content more than enough in Certification Exam point of view. 

Before going in to the topic, let us know what is Model Evaluation. It is the process of  utilizing the various evaluation metrics to understand the Machine Learning Model's Performance. So now lets get started,

Model Evaluation - Regression Model:

    You will be getting at least one question from this topic in AI-900 Certification Exam. So keep below points as thumb rule for this topic.

    i) Mean Absolute Error (MAE) : 

        It answers the question that How close is a prediction to actual value?. Lower the value better the result.

    ii) Mean Squared Error (MSE) :

        It is the Average of Square of distance between actual and predicted values. When you want to penalize large prediction errors  then go for Mean Squared Error. Example for Mean Squared Error Scenario usage is "House Value Prediction".  Lower the value better the result.

    iii) Root Mean Squared Error (RMSE):

        It is the Square root of Mean Squared Error that is mentioned above. It is the alternative of Mean Squared Error. Lower the value better the result.


Model Evaluation - Classification Model:

    Terminology used here are Predicted label, Truce label and Confusion Matrix. Let us see one by one.

    i) Predicted Label:

        It is nothing but the predicted value that is output of the model.

    ii) True Label:

        It is the expected value. That is we are expecting the value this must be the result for our created model.

    iii) Confusion Matrix:

        It is nothing but, a Matrix which matches the predicted label vs true label.


        


    Above matrix is called Confusion Matrix. It will looks little bit confusion as its name 😜. Don't get confused. To make it simple , take prediction as x axis and Actual value as Y axis. Note that,

    P represents Positive
    N represents Negative
    TN represents True Negative
    FP represents False Positive
    FN represents False Negative
    TP represents True Positive.

Just giving a simple words on the creation of value of True Negative and that is same for all others. If the Actual Value is Negative and the Predicted value is also Negative then we can say the answer is correct, that is our Actual and Prediction value are same so it is "True" and we need to add the predicted value at last and that is "Negative". So on joining the both we get "True Negative".

    For perfect Model we need, True Negative and False Negative with higher value along with this False Positive and False Negative as 0. But in real scenario , different use cases have different needs.


Model Evaluation metrics for classification:

    These are mathematics based on confusion matrix, so don't need to deep dive into the mathematics or formula part on this topic in certification point of view. But there will be question on this topic, so better know the what are the metric are there. This will be more than enough.

    i) Accuracy:

        (TP+TN)/Total Number of Cases

    ii) Precession :

        TP/(TP+FP)

    iii) Recall:

        TP/(TP+FN)

    iv) F1 Score:

        2TP/(2TP+FP+FN)


Thus in this blog we saw about Model Evaluation of Regression and classification Model and their Metrix . This is purely on the point of view of AI-900 Certification Exam. Hope this information is helpful.

Thank You !!!

'