In Azure Machine Learning, it’s possible to create, train, and deploy machine learning models. But to ensure these models are accurate and reliable, a critical step in the process is performance evaluation. Our goal will be to show how you can evaluate models in Azure Machine Learning.

#### Creating and Evaluating Models

We’ll divide the process into two parts: model creation and model evaluation. For model creation, you build your model by feeding data into a supervised or unsupervised learning algorithm. Once the model is created, you evaluate it using specific metrics to ensure the model is providing accurate predictions.

#### Evaluating Classification Models

Classification models predict categorical outcomes. For example, whether an email is spam (yes or no), or classifying images of fruits into their respective categories (apples, oranges, bananas, etc). Evaluation techniques include confusion matrix, accuracy, precision, recall, AUC, and F1 score. Let’s take Confusion Matrix as an example:

A confusion matrix is a table that is often used to describe the performance of a classification model on a set of test data for which the true values are known. The matrix includes 4 distinct categories:

- True Positives (TP): The model correctly predicted the positive class
- True Negatives (TN): The model correctly predicted the negative class
- False Positives (FP): The model incorrectly predicted the positive class
- False Negatives (FN): The model incorrectly predicted the negative class

#### Evaluating Regression Models

Regression models predict continuous outcomes, like predicting housing prices or stock market trends. Evaluation techniques include Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE) and the Coefficient of Determination (R-Squared). Let’s consider Mean Absolute Error:

Mean Absolute Error represents the average of the absolute differences between predictions and actual values. Lower values of MAE indicate better fit of data.

MAE = Average (|True Value – Predicted Value|)

#### Evaluating Clustering Models

Clustering models divide data into groups or clusters based on similarities. These models are evaluated using methods such as silhouette score, Davies-Bouldin index, and Rand index, among others. As an example, the Silhouette score measures how close each point in one cluster is to the points in the neighboring clusters.

### Final Words

Evaluating models is an integral part of the machine learning workflow. Appropriate evaluation ensures the built models are accurate and reliable. The DP-100 exam includes multiple questions on model evaluation, thus, mastering these concepts is necessary for success. Leveraging Azure Machine Learning for model evaluation provides the benefit of easily testing metrics and visualizing results. It also supports re-use and sharing of analysis pipelines and models across your organization to ensure effective and efficient collaboration.

## Practice Test

### True or False: The metric ‘Root Mean Squared Error (RMSE)’ is used to evaluate the performance of a classification model.

- True
- False

**Answer:** False

**Explanation:** RMSE is used in regression analysis to measure the differences between the predicted or estimated values and the actual values.

### What model evaluation metric measures the percentage of the Total Sum of Squares that is ‘explained’ by the regression?

- A. AUC-ROC
- B. Mean Square Error
- C. R-squared
- D. Precision

**Answer:** C. R-squared

**Explanation:** R-squared is the measure of how close the data are to the fitted regression line or it explains the percentage of the total sum of squares that is ‘explained’ by the regression.

### True or False: Precision-Recall is a useful metric of success for some imbalanced classification problems.

- True
- False

**Answer:** True

**Explanation:** Precision-Recall is indeed a good metric for imbalanced classification problems, as it is more concerned with the positive class being predicted correctly.

### A _________ is an error from a model predicting values that are excessively far from their actual values.

- A. Overfitting
- B. Underfitting
- C. Outlier
- D. Residual

**Answer:** C. Outlier

**Explanation:** Outliers refer to values that are significantly different or distant from the other values predicted by the model which can distort the model’s performance.

### Which of the following models is more prone to overfitting?

- A. Simple model
- B. Complex model
- C. Low variance model
- D. High bias model

**Answer:** B. Complex model

**Explanation:** Overfitting typically happens when the model is too complex. It learns the detail and noise in the training data which negatively impact the performance of the model on the new data.

### True or False: A perfect model has an AUC-ROC score of

- True
- False

**Answer:** True

**Explanation:** The AUC-ROC score of a perfect model is An area of 1 represents a perfect test; an area of .5 represents a worthless test.

### The F1 score is the harmonic mean of precision and ________.

- A. R-squared
- B. Recall
- C. Root Mean Square Error
- D. Accuracy

**Answer:** B. Recall

**Explanation:** The F1 score is the harmonic mean of precision and recall, providing a balance between these two metrics.

### Which model metrics would you consider for a multi-class classification problem?

- A. AUC-ROC
- B. Accuracy
- C. Recall
- D. All of the above

**Answer:** D. All of the above

**Explanation:** All these metrics could be relevant in a multi-class classification problem.

### True or False: Underfitting occurs when a model is too simple to capture the underlying pattern of the data.

- True
- False

**Answer:** True

**Explanation:** Underfitting happens when a model cannot aptly capture the underlying structure of the data.

### Cross-validation technique is used for ________.

- A. Model Selection
- B. Feature selection
- C. Model Evaluation
- D. Both A and C

**Answer:** D. Both A and C

**Explanation:** Cross-validation can be used both for model selection and model evaluation by partitioning the available data and using some partitions for training and some for testing.

### For imbalanced classification problem, which of the following metrics you should not rely on?

- A. Precision
- B. Recall
- C. F1 Score
- D. Accuracy

**Answer:** D. Accuracy

**Explanation:** Accuracy can be misleading for imbalanced classification, as a model that only predicts the majority class could still have high accuracy.

### True or False: In general, a lower value of RMSE signifies a better fit of the model.

- True
- False

**Answer:** True

**Explanation:** RMSE measures the difference between the predictions of a model and the observed data. Lower RMSE values indicate better fit – as the differences between predicted and observed values are smaller.

### In confusion matrix, what does TP stand for?

- A. True precision
- B. True Positive
- C. Total probability
- D. Total precision

**Answer:** B. True Positive

**Explanation:** True Positive (TP) refers to the number of positives that have been correctly identified by the model.

### Which two model evaluation metrics can be combined into the F1 Score?

- A. Accuracy and Precision
- B. Precision and Recall
- C. Accuracy and Recall
- D. Precision and Negative Predictive Value

**Answer:** B. Precision and Recall

**Explanation:** F1 Score is the harmonic mean of Precision and Recall and acts as a better measure of the incorrectly classified cases than the Accuracy Metric.

### True or False: Overfitting happens when the model performs well on the testing data but poorly on the training data.

- True
- False

**Answer:** False

**Explanation:** Overfitting refers to the situation where model performs well on the training data but poorly on the testing data as it tends to capture noise along with the underlying pattern in the data.

## Interview Questions

### What is model evaluation in data science?

Model evaluation is the process of assessing the predictive performance, reliability, and robustness of a trained predictive model. Typically this involves comparing the model’s predictions with actual outcomes to check its reliability and performance.

### In Azure ML, how can you evaluate a model’s performance?

In Azure ML, you can use evaluation metrics provided by the Azure Machine Learning designer. These metrics are used to measure model performance and understand its accuracy.

### What are some common metrics used to evaluate the performance of a regression model in Azure Machine Learning?

Some common metrics used to evaluate the regression models in Azure Machine Learning include Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Relative Absolute Error (RAE), and Relative Squared Error (RSE).

### What is Cross-Validation in Azure machine learning?

Cross-validation is a model validation technique in Azure Machine Learning that divides data into two sections: one for training a model and the other for validating a model. This process helps to assess how the results of a statistical analysis will generalize to an independent data set.

### What is a confusion matrix in model evaluation?

A confusion matrix is a tabular representation of Actual vs Predicted values. It helps us to see the performance of the prediction and the types of errors made by the model, including true positives, true negatives, false positives, and false negatives.

### What is AUC-ROC in model evaluation?

AUC-ROC (Area Under the Receiver Operating Characteristic) is a performance measurement for the classification problems at various thresholds settings. It tells us how much the model is capable of distinguishing between classes.

### How can you perform model evaluation in Azure Machine Learning Studio?

You can evaluate your model in Azure ML Studio by first training your model using the ‘Train Model’ or ‘Tune Model Hyperparameters’ module. Following this step, the ‘Score Model’ module is used to generate predictions. Lastly, the ‘Evaluate Model’ module compares these predictions to true values to assess the performance of the model.

### What does Mean Average Precision (MAP) mean in model evaluation?

Mean Average Precision (MAP) is a measure that combines precision and recall into a single value. It is particularly useful in information retrieval and ranking problems, where for a single query, multiple possible answers exist.

### What is R2 (coefficient of determination) in model evaluation?

The R2 or the coefficient of determination is a statistical metric that provides a measure of how well future outcomes are likely to be predicted by the model. R2 value near 1 indicates a good fit of the model.

### How can you perform modeling and evaluation with Python SDK in Azure machine learning?

The Azure Machine Learning Python SDK allows you to train and evaluate the model using familiar Python libraries. You would use the Azure ML SDK to submit training jobs and retrieve the results, including any saved models. Evaluation can be performed using typical package-specific methods or Azure ML’s model interpretability features.