Metrics To Evaluate Machine Learning Algorithms in Python

 The metrics that you choose to evaluate your machine learning algorithms are very important.

Choice of metrics influences how the performance of machine learning algorithms is measured and compared. They influence how you weight the importance of different characteristics in the results and your ultimate choice of which algorithm to choose.

In this post, you will discover how to select and use different machine learning performance metrics in Python with scikit-learn.

About the Recipes

Various different machine learning evaluation metrics are demonstrated in this post using small code recipes in Python and scikit-learn.

Each recipe is designed to be standalone so that you can copy-and-paste it into your project and use it immediately.

Metrics are demonstrated for both classification and regression type machine learning problems.

• For classification metrics, the Pima Indians onset of diabetes dataset is used as demonstration. This is a binary classification problem where all of the input variables are numeric (update: download from here).
• For regression metrics, the Boston House Price dataset is used as demonstration. this is a regression problem where all of the input variables are also numeric (update: download data from here).

In each recipe, the dataset is downloaded directly.

All recipes evaluate the same algorithms, Logistic Regression for classification and Linear Regression for the regression problems. A 10-fold cross-validation test harness is used to demonstrate each metric, because this is the most likely scenario where you will be employing different algorithm evaluation metrics.

A caveat in these recipes is the cross_val_score function used to report the performance in each recipe.It does allow the use of different scoring metrics that will be discussed, but all scores are reported so that they can be sorted in ascending order (largest score is best).

Some evaluation metrics (like mean squared error) are naturally descending scores (the smallest score is best) and as such are reported as negative by the cross_val_score() function. This is important to note, because some scores will be reported as negative that by definition can never be negative.

You can learn more about machine learning algorithm performance metrics supported by scikit-learn on the page Model evaluation: quantifying the quality of predictions.

Let’s get on with the evaluation metrics.

Need help with Machine Learning in Python?

Take my free 2-week email course and discover data prep, algorithms and more (with code).

Click to sign-up now and also get a free PDF Ebook version of the course.

Classification Metrics

Classification problems are perhaps the most common type of machine learning problem and as such there are a myriad of metrics that can be used to evaluate predictions for these problems.

In this section we will review how to use the following metrics:

1. Classification Accuracy.
2. Log Loss.
3. Area Under ROC Curve.
4. Confusion Matrix.
5. Classification Report.

1. Classification Accuracy

Classification accuracy is the number of correct predictions made as a ratio of all predictions made.

This is the most common evaluation metric for classification problems, it is also the most misused. It is really only suitable when there are an equal number of observations in each class (which is rarely the case) and that all predictions and prediction errors are equally important, which is often not the case.

Below is an example of calculating classification accuracy.

# Cross Validation Classification Accuracy import pandas from sklearn import model_selection from sklearn.linear_model import LogisticRegression url = "https://raw.githubusercontent.com/jbrownlee/Datasets/master/pima-indians-diabetes.data.csv" names = ['preg', 'plas', 'pres', 'skin', 'test', 'mass', 'pedi', 'age', 'class'] dataframe = pandas.read_csv(url, names=names) array = dataframe.values X = array[:,0:8] Y = array[:,8] kfold = model_selection.KFold(n_splits=10, random_state=7, shuffle=True) model = LogisticRegression(solver='liblinear') scoring = 'accuracy' results = model_selection.cross_val_score(model, X, Y, cv=kfold, scoring=scoring) print("Accuracy: %.3f (%.3f)" % (results.mean(), results.std()))

You can see that the ratio is reported.

Note: Your results may vary given the stochastic nature of the algorithm or evaluation procedure, or differences in numerical precision. Consider running the example a few times and compare the average outcome.

This can be converted into a percentage by multiplying the value by 100, giving an accuracy score of approximately 77% accurate.

Accuracy: 0.770 (0.048)

2. Log Loss

Logistic loss (or log loss) is a performance metric for evaluating the predictions of probabilities of membership to a given class.

The scalar probability between 0 and 1 can be seen as a measure of confidence for a prediction by an algorithm. Predictions that are correct or incorrect are rewarded or punished proportionally to the confidence of the prediction.

For more on log loss and it’s relationship to cross-entropy, see the tutorial:

• A Gentle Introduction to Cross-Entropy for Machine Learning

Below is an example of calculating log loss for Logistic regression predictions on the Pima Indians onset of diabetes dataset.

# Cross Validation Classification LogLoss import pandas from sklearn import model_selection from sklearn.linear_model import LogisticRegression url = "https://raw.githubusercontent.com/jbrownlee/Datasets/master/pima-indians-diabetes.data.csv" names = ['preg', 'plas', 'pres', 'skin', 'test', 'mass', 'pedi', 'age', 'class'] dataframe = pandas.read_csv(url, names=names) array = dataframe.values X = array[:,0:8] Y = array[:,8] kfold = model_selection.KFold(n_splits=10, random_state=7, shuffle=True) model = LogisticRegression(solver='liblinear') scoring = 'neg_log_loss' results = model_selection.cross_val_score(model, X, Y, cv=kfold, scoring=scoring) print("Logloss: %.3f (%.3f)" % (results.mean(), results.std()))

Smaller log loss is better with 0 representing a perfect log loss.

Note: Your results may vary given the stochastic nature of the algorithm or evaluation procedure, or differences in numerical precision. Consider running the example a few times and compare the average outcome.

As mentioned above, the measure is inverted to be ascending when using the cross_val_score() function.

Logloss: -0.493 (0.047)

3. Area Under ROC Curve

Area Under ROC Curve (or ROC AUC for short) is a performance metric for binary classification problems.

The AUC represents a model’s ability to discriminate between positive and negative classes. An area of 1.0 represents a model that made all predictions perfectly. An area of 0.5 represents a model as good as random.

A ROC Curve is a plot of the true positive rate and the false positive rate for a given set of probability predictions at different thresholds used to map the probabilities to class labels. The area under the curve is then the approximate integral under the ROC Curve.

For more on ROC Curves and ROC AUC, see the tutorial:

• How to Use ROC Curves and Precision-Recall Curves for Classification in Python

The example below provides a demonstration of calculating AUC.

# Cross Validation Classification ROC AUC import pandas from sklearn import model_selection from sklearn.linear_model import LogisticRegression url = "https://raw.githubusercontent.com/jbrownlee/Datasets/master/pima-indians-diabetes.data.csv" names = ['preg', 'plas', 'pres', 'skin', 'test', 'mass', 'pedi', 'age', 'class'] dataframe = pandas.read_csv(url, names=names) array = dataframe.values X = array[:,0:8] Y = array[:,8] kfold = model_selection.KFold(n_splits=10, random_state=7, shuffle=True) model = LogisticRegression(solver='liblinear') scoring = 'roc_auc' results = model_selection.cross_val_score(model, X, Y, cv=kfold, scoring=scoring) print("AUC: %.3f (%.3f)" % (results.mean(), results.std()))

Note: Your results may vary given the stochastic nature of the algorithm or evaluation procedure, or differences in numerical precision. Consider running the example a few times and compare the average outcome.

You can see the the AUC is relatively close to 1 and greater than 0.5, suggesting some skill in the predictions.

AUC: 0.824 (0.041)

4. Confusion Matrix

The confusion matrix is a handy presentation of the accuracy of a model with two or more classes.

The table presents predictions on the x-axis and accuracy outcomes on the y-axis. The cells of the table are the number of predictions made by a machine learning algorithm.

For example, a machine learning algorithm can predict 0 or 1 and each prediction may actually have been a 0 or 1. Predictions for 0 that were actually 0 appear in the cell for prediction=0 and actual=0, whereas predictions for 0 that were actually 1 appear in the cell for prediction = 0 and actual=1. And so on.

For more on the confusion matrix, see this tutorial:

• What is a Confusion Matrix in Machine Learning

Below is an example of calculating a confusion matrix for a set of prediction by a model on a test set.

# Cross Validation Classification Confusion Matrix import pandas from sklearn import model_selection from sklearn.linear_model import LogisticRegression from sklearn.metrics import confusion_matrix url = "https://raw.githubusercontent.com/jbrownlee/Datasets/master/pima-indians-diabetes.data.csv" names = ['preg', 'plas', 'pres', 'skin', 'test', 'mass', 'pedi', 'age', 'class'] dataframe = pandas.read_csv(url, names=names) array = dataframe.values X = array[:,0:8] Y = array[:,8] test_size = 0.33 X_train, X_test, Y_train, Y_test = model_selection.train_test_split(X, Y, test_size=test_size, random_state=7) model = LogisticRegression(solver='liblinear') model.fit(X_train, Y_train) predicted = model.predict(X_test) matrix = confusion_matrix(Y_test, predicted) print(matrix)

Note: Your results may vary given the stochastic nature of the algorithm or evaluation procedure, or differences in numerical precision. Consider running the example a few times and compare the average outcome.

Although the array is printed without headings, you can see that the majority of the predictions fall on the diagonal line of the matrix (which are correct predictions).

[[141  21] [ 41  51]]

5. Classification Report

Scikit-learn does provide a convenience report when working on classification problems to give you a quick idea of the accuracy of a model using a number of measures.

The classification_report() function displays the precision, recall, f1-score and support for each class.

The example below demonstrates the report on the binary classification problem.

# Cross Validation Classification Report import pandas from sklearn import model_selection from sklearn.linear_model import LogisticRegression from sklearn.metrics import classification_report url = "https://raw.githubusercontent.com/jbrownlee/Datasets/master/pima-indians-diabetes.data.csv" names = ['preg', 'plas', 'pres', 'skin', 'test', 'mass', 'pedi', 'age', 'class'] dataframe = pandas.read_csv(url, names=names) array = dataframe.values X = array[:,0:8] Y = array[:,8] test_size = 0.33 X_train, X_test, Y_train, Y_test = model_selection.train_test_split(X, Y, test_size=test_size, random_state=7) model = LogisticRegression(solver='liblinear') model.fit(X_train, Y_train) predicted = model.predict(X_test) report = classification_report(Y_test, predicted) print(report)

Note: Your results may vary given the stochastic nature of the algorithm or evaluation procedure, or differences in numerical precision. Consider running the example a few times and compare the average outcome.

You can see good prediction and recall for the algorithm.

precision    recall  f1-score   support           0.0       0.77      0.87      0.82       162         1.0       0.71      0.55      0.62        92   avg / total       0.75      0.76      0.75       254

Regression Metrics

In this section will review 3 of the most common metrics for evaluating predictions on regression machine learning problems:

1. Mean Absolute Error.
2. Mean Squared Error.
3. R^2.

1. Mean Absolute Error

The Mean Absolute Error (or MAE) is the average of the absolute differences between predictions and actual values. It gives an idea of how wrong the predictions were.

The measure gives an idea of the magnitude of the error, but no idea of the direction (e.g. over or under predicting).

You can learn more about Mean Absolute error on Wikipedia.

The example below demonstrates calculating mean absolute error on the Boston house price dataset.

# Cross Validation Regression MAE import pandas from sklearn import model_selection from sklearn.linear_model import LinearRegression url = "https://raw.githubusercontent.com/jbrownlee/Datasets/master/housing.data" names = ['CRIM', 'ZN', 'INDUS', 'CHAS', 'NOX', 'RM', 'AGE', 'DIS', 'RAD', 'TAX', 'PTRATIO', 'B', 'LSTAT', 'MEDV'] dataframe = pandas.read_csv(url, delim_whitespace=True, names=names) array = dataframe.values X = array[:,0:13] Y = array[:,13] kfold = model_selection.KFold(n_splits=10, random_state=7, shuffle=True) model = LinearRegression() scoring = 'neg_mean_absolute_error' results = model_selection.cross_val_score(model, X, Y, cv=kfold, scoring=scoring) print("MAE: %.3f (%.3f)" % (results.mean(), results.std()))

A value of 0 indicates no error or perfect predictions.

Note: Your results may vary given the stochastic nature of the algorithm or evaluation procedure, or differences in numerical precision. Consider running the example a few times and compare the average outcome.

Like logloss, this metric is inverted by the cross_val_score() function.

MAE: -4.005 (2.084)

2. Mean Squared Error

The Mean Squared Error (or MSE) is much like the mean absolute error in that it provides a gross idea of the magnitude of error.

Taking the square root of the mean squared error converts the units back to the original units of the output variable and can be meaningful for description and presentation. This is called the Root Mean Squared Error (or RMSE).

You can learn more about Mean Squared Error on Wikipedia.

The example below provides a demonstration of calculating mean squared error.

# Cross Validation Regression MSE import pandas from sklearn import model_selection from sklearn.linear_model import LinearRegression url = "https://raw.githubusercontent.com/jbrownlee/Datasets/master/housing.data" names = ['CRIM', 'ZN', 'INDUS', 'CHAS', 'NOX', 'RM', 'AGE', 'DIS', 'RAD', 'TAX', 'PTRATIO', 'B', 'LSTAT', 'MEDV'] dataframe = pandas.read_csv(url, delim_whitespace=True, names=names) array = dataframe.values X = array[:,0:13] Y = array[:,13] kfold = model_selection.KFold(n_splits=10, random_state=7, shuffle=True) model = LinearRegression() scoring = 'neg_mean_squared_error' results = model_selection.cross_val_score(model, X, Y, cv=kfold, scoring=scoring) print("MSE: %.3f (%.3f)" % (results.mean(), results.std()))

This metric too is inverted so that the results are increasing.

Note: Your results may vary given the stochastic nature of the algorithm or evaluation procedure, or differences in numerical precision. Consider running the example a few times and compare the average outcome.

Remember to take the absolute value before taking the square root if you are interested in calculating the RMSE.

MSE: -34.705 (45.574)

3. R^2 Metric

The R^2 (or R Squared) metric provides an indication of the goodness of fit of a set of predictions to the actual values. In statistical literature, this measure is called the coefficient of determination.

This is a value between 0 and 1 for no-fit and perfect fit respectively.

You can learn more about the Coefficient of determination article on Wikipedia.

The example below provides a demonstration of calculating the mean R^2 for a set of predictions.

# Cross Validation Regression R^2 import pandas from sklearn import model_selection from sklearn.linear_model import LinearRegression url = "https://raw.githubusercontent.com/jbrownlee/Datasets/master/housing.data" names = ['CRIM', 'ZN', 'INDUS', 'CHAS', 'NOX', 'RM', 'AGE', 'DIS', 'RAD', 'TAX', 'PTRATIO', 'B', 'LSTAT', 'MEDV'] dataframe = pandas.read_csv(url, delim_whitespace=True, names=names) array = dataframe.values X = array[:,0:13] Y = array[:,13] kfold = model_selection.KFold(n_splits=10, random_state=7, shuffle=True) model = LinearRegression() scoring = 'r2' results = model_selection.cross_val_score(model, X, Y, cv=kfold, scoring=scoring) print("R^2: %.3f (%.3f)" % (results.mean(), results.std()))

Note: Your results may vary given the stochastic nature of the algorithm or evaluation procedure, or differences in numerical precision. Consider running the example a few times and compare the average outcome.

You can see that the predictions have a poor fit to the actual values with a value close to zero and less than 0.5.

R^2: 0.203 (0.595)

Summary

In this post, you discovered metrics that you can use to evaluate your machine learning algorithms.

You learned about 3 classification metrics:

• Accuracy.
• Log Loss.
• Area Under ROC Curve.

Also 2 convenience methods for classification prediction results:

• Confusion Matrix.
• Classification Report.

And 3 regression metrics:

• Mean Absolute Error.
• Mean Squared Error.
• R^2.

Do you have any questions about metrics for evaluating machine learning algorithms or this post? Ask your question in the comments and I will do my best to answer it.