Neural Networks

Neural networks for binary and multiclass classification

Neural network models are structured as a series of layers that reflect the way the brain processes information. The neural network classifiers available in Statistics and Machine Learning Toolbox™ are fully connected, feedforward neural networks for which you can adjust the size of the fully connected layers and change the activation functions of the layers.

To train a neural network classification model, use the Classification Learner app. The app offers several types of fully connected networks. If you have Deep Learning Toolbox™, you can also edit and train customizable neural network classifiers. For more information, see Choose Classifier Options in Classification Learner.

For greater flexibility, train a neural network classifier using fitcnet in the command-line interface. After training, you can classify new data by passing the model and the new predictor data to predict.

If you want to create more complex deep learning networks and have Deep Learning Toolbox, you can try the Deep Network Designer (Deep Learning Toolbox) app.

Apps

Classification Learner

Train models to classify data using supervised machine learning

Blocks

ClassificationNeuralNetwork Predict

Classify observations using neural network classification model (Since R2021b)

Functions

expand all

Create Neural Network Model

`fitcnet`	Train neural network classification model
`compact`	Reduce size of machine learning model

Interpret Prediction

`lime`	Local interpretable model-agnostic explanations (LIME)
`partialDependence`	Compute partial dependence
`permutationImportance`	Predictor importance by permutation (Since R2024a)
`plotPartialDependence`	Create partial dependence plot (PDP) and individual conditional expectation (ICE) plots
`shapley`	Shapley values

Cross-Validate

`crossval`	Cross-validate machine learning model
`kfoldLoss`	Classification loss for cross-validated classification model
`kfoldPredict`	Classify observations in cross-validated classification model
`kfoldEdge`	Classification edge for cross-validated classification model
`kfoldMargin`	Classification margins for cross-validated classification model
`kfoldfun`	Cross-validate function for classification

Measure Performance

`loss`	Classification loss for neural network classifier
`resubLoss`	Resubstitution classification loss
`edge`	Classification edge for neural network classifier
`margin`	Classification margins for neural network classifier
`resubEdge`	Resubstitution classification edge
`resubMargin`	Resubstitution classification margin

Classify Observations

`predict`	Classify observations using neural network classifier
`resubPredict`	Classify training data using trained classifier

Objects

`ClassificationNeuralNetwork`	Neural network model for classification
`CompactClassificationNeuralNetwork`	Compact neural network model for classification
`ClassificationPartitionedNeuralNetwork`	Cross-validated classification neural network (Since R2026a)

Topics

Assess Neural Network Classifier Performance
Use fitcnet to create a feedforward neural network classifier with fully connected layers, and assess the performance of the model on test data.
Compress Machine Learning Model for Memory-Limited Hardware
Reduce model size by feature selection, constrained Bayesian optimization, and parameter quantization.