Definition
An artificial neural network is a model built from layers of connected "neurons" that learns from data to approximate even very complex nonlinear relationships between inputs and an output. Multilayer perceptrons (MLP) and radial basis function (RBF) networks serve classification, regression, time-series forecasting and clustering; the network is trained by repeatedly adjusting weights to minimize error.
Data mining & machine learning
Networks excel where the relationship is nonlinear and data are plentiful; the price is poorer interpretability and the need to guard against overfitting with a test set and early stopping. Inputs must be scaled and categorical variables encoded.
In practice networks are used to predict consumption, output quality from process parameters, credit risk or campaign response. It pays to compare them with tree-based methods — boosting often wins on tabular data, networks on signals and time series.
In Statistica
The Statistica Automated Neural Networks (SANN) module in the Data Mining menu automatically designs and trains dozens of MLP and RBF architectures, picks the best by test error, handles classification, regression, time series and clustering (Kohonen maps), offers input sensitivity analysis and saves the network as code (C, PMML) for deployment.
Related terms
- Random forests and boostingRandom forests and boosting are ensemble methods that combine hundreds of decision trees into one more accurate model.
- Nonlinear regressionNonlinear regression fits data with a model that is nonlinear in its parameters — exponential growth, a logistic curve…
- OverfittingOverfitting happens when a model learns not only the real relationships but also the random noise of the training data…
- Cross-validationCross-validation estimates how well a model will perform on new data.
Knowledgebase guides
FAQ
- How much data does a neural network need?
- Hundreds to thousands of observations depending on the complexity of the relationship; for small datasets regression or trees give more reliable results.
- How do I know the network is overfitted?
- Training error keeps falling while test error rises. Use early stopping, a smaller network or regularization.
Try it on your own data
Statistica free for 30 days
Full version, no credit card. Or get a pricing estimate in a minute.
Data mining & machine learning
Updated: September 2026.