This is machine translation

Translated by Microsoft
Mouseover text to see original. Click the button below to return to the English version of the page.

Note: This page has been translated by MathWorks. Click here to see
To view all translated materials including this page, select Country from the country navigator on the bottom of this page.

Select Data and Validation for Classification Problem

Select Data from the Workspace

Tip

In Classification Learner, tables are the easiest way to use your data, because they can contain numeric and label data. Use the Import Tool to bring your data into the MATLAB® workspace as a table, or use the table functions to create a table from workspace variables. See Tables (MATLAB).

If your predictors are a matrix and the response is a vector, combine them into a table using the table function.

  1. Load your data into the MATLAB workspace.

    Predictor and response variables can be numeric, categorical, string, or logical vectors, cell arrays of character vectors, or character arrays. If your response variable is a string vector, then the predictions of the trained model is a cell array of character vectors.

    For example data sets, see Example Data for Classification.

  2. On the Apps tab, click Classification Learner.

  3. In Classification Learner, on the Classification Learner tab, in the File section, click New Session.

  4. In the New Session dialog box, select a table or matrix from the workspace variables.

    If you select a matrix, choose whether to use rows or columns for observations by clicking the option buttons.

  5. Observe the roles the app selects for the variables based on their data type. The app tries to select a suitable response variable, and all other variables are predictors. Change the selections if needed. Add or remove predictors using the check boxes. Add or remove all predictors by clicking Add All or Remove All. You can also add or remove multiple predictors by selecting them in the table, and then clicking Add N or Remove N, where N is the number of selected predictors. The Add All and Remove All buttons change to Add N and Remove N when you select multiple predictors.

  6. To accept the default validation scheme and continue, click Start Session. The default validation option is 5-fold cross-validation, which protects against overfitting.

    Tip

    If you have a large data set you might want to switch to holdout validation. To learn more, see Choose Validation Scheme.

For next steps, see Train Classification Models in Classification Learner App.

Import Data from File

  1. On the Classification Learner tab, in the File section, select New Session > From File.

  2. Select a file type in the list, such as spreadsheets, text files, or comma separated values (.csv) files, or select All Files to browse for other file types such as .dat.

Example Data for Classification

To get started using Classification Learner, try the following example data sets.

NameSizeDescription
Fisher Iris

Number of predictors: 4
Number of observations: 150
Number of classes: 3
Response: species

Measurements from three species of iris. Try to classify the species.

For a step-by-step example, see Train Decision Trees Using Classification Learner App.

Create a table from the .csv file:

fishertable = readtable('fisheriris.csv');

Credit Rating

Number of predictors: 6
Number of observations: 3932
Number of classes: 7
Response: Rating

Financial ratios and industry sectors information for a list of corporate customers. The response variable consists of credit ratings (AAA, AA, A, BBB, BB, B, CCC) assigned by a rating agency.

Create a table from the CreditRating_Historical.dat file:

creditrating = readtable('CreditRating_Historical.dat');

Cars

Number of predictors: 7
Number of observations: 100
Number of classes: 7
Response: Origin

Measurements of cars, in 1970, 1976, and 1982. Try to classify the country of origin.

Create a table from variables in the carsmall.mat file:

load carsmall
cartable = table(Acceleration, Cylinders, Displacement,...
Horsepower, Model_Year, MPG, Weight, Origin);

Arrhythmia

Number of predictors: 279
Number of observations: 452
Number of classes: 16
Response: Class (Y)

Patient information and response variables that indicate the presence and absence of cardiac arrhythmia. Misclassifying a patient as "normal" has more severe consequences than false positives classified as “has arrhythmia”.

Create a table from the .mat file:

load arrhythmia
Arrhythmia = array2table(X);
Arrhythmia.Class = categorical(Y);

Ovarian Cancer

Number of predictors: 4000
Number of observations: 216
Number of classes: 2
Response: Group

Ovarian cancer data generated using the WCX2 protein array. Includes 95 controls and 121 ovarian cancers.

Create a table from the .mat file:

load ovariancancer
ovariancancer = array2table(obs);
ovariancancer.Group = categorical(grp);

Ionosphere

Number of predictors: 34
Number of observations: 351
Number of classes: 2
Response: Group (Y)

Signals from a phased array of 16 high-frequency antennas. Good (“g”) returned radar signals are those showing evidence of some type of structure in the ionosphere. Bad (“b”) signals are those that pass through the ionosphere.

Create a table from the .mat file:

load ionosphere
ionosphere = array2table(X);
ionosphere.Group = Y;

Choose Validation Scheme

Choose a validation method to examine the predictive accuracy of the fitted models. Validation estimates model performance on new data compared to the training data, and helps you choose the best model. Validation protects against overfitting. Choose a validation scheme before training any models, so that you can compare all the models in your session using the same validation scheme.

Tip

Try the default validation scheme and click Start Session to continue. The default option is 5-fold cross-validation, which protects against overfitting.

If you have a large data set and training models takes too long using cross-validation, reimport your data and try the faster holdout validation instead.

  • Cross-Validation: Select a number of folds (or divisions) to partition the data set using the slider control.

    If you choose k folds, then the app:

    1. Partitions the data into k disjoint sets or folds

    2. For each fold:

      1. Trains a model using the out-of-fold observations

      2. Assesses model performance using in-fold data

    3. Calculates the average test error over all folds

    This method gives a good estimate of the predictive accuracy of the final model trained with all the data. It requires multiple fits but makes efficient use of all the data, so it is recommended for small data sets.

  • Holdout Validation: Select a percentage of the data to use as a test set using the slider control. The app trains a model on the training set and assesses its performance with the test set. The model used for validation is based on only a portion of the data, so Holdout Validation is recommended only for large data sets. The final model is trained with the full data set.

  • No Validation: No protection against overfitting. The app uses all of the data for training and computes the error rate on the same data. Without any test data, you get an unrealistic estimate of the model’s performance on new data. That is, the training sample accuracy is likely to be unrealistically high, and the predictive accuracy is likely to be lower.

    To help you avoid overfitting to the training data, choose a validation scheme instead.

Note

The validation scheme only affects the way that Classification Learner computes validation metrics. The final model is always trained using the full data set.

All the classification models you train after selecting data use the same validation scheme that you select in this dialog box. You can compare all the models in your session using the same validation scheme.

To change the validation selection and train new models, you can select data again, but you lose any trained models. The app warns you that importing data starts a new session. Save any trained models you want to keep to the workspace, and then import the data.

For next steps training models, see Train Classification Models in Classification Learner App.

Related Topics