Home

WiSARD4WEKA

A supervised classification model for WEKA based on the WiSARD weightless neural network for the Weka machine learning toolkit.

WiSARD was originally conceived as a pattern recognition device mainly focusing on image processing domain. With ad hoc data transformation, WiSARD can also be used successfully as multiclass classifier in machine learning domain.

The WiSARD is a RAM-based neuron network working as an n-tuple classifier. A WiSARD is formed by as many discriminators as the number of classes it has to discriminate between. Each discriminator consists of a set of N RAMs that, during the training phase, l earn the occurrences of n-tuples extracted from the input binary vector (the retina).

In the WiSARD model, n-tuples selected from the input binary vector are regarded as the “features” of the input pattern to be recognized. It has been demostrated in literature [14] that the randomness of feature extraction makes WiSARD more sensitive to detect global features than an ordered map which makes a single layer system sensitive to detect local features.

More information and details about the WiSARD neural network model can be found in Aleksander and Morton's book Introduction to neural computing.

The WiSARD4WEKA package implements a multi-class classification method based on the WiSARD weightless neural model for the Weka machine learning toolkit. A data-preprocessing filter allows to exploit WiSARD neural model training/classification capabilities on multi-attribute numeric data making WiSARD overcome the restriction to binary pattern recognition.

For more information on the WiSARD classifier implemented in the WiSARD4WEKA package, see:

Massimo De Gregorio and Maurizio Giordano (2018).
An experimental evaluation of weightless neural networks for multi-class classification.
Journal of Applied Soft Computing. Vol.72. pp. 338-354