Classification of Paper Quality Using Neural Networks:
Introduction

In this example different neural classifiers are compared using data from a hybrid gas array sensor, an electronic nose. The odors from five different cardboard papers from commercial manufacturers are recorded with the electronic nose. Different kinds of classifiers are trained to determine the origin of an unknown sample.

The data was kindly contributed by the Swedish Sensor Centre, S-SENCE, http://www.ifm.liu.se/Applphys/S-SENCE. More information on the data set can be found in "Identification of Paper Quality Using a Hybrid Electronic Nose" by Martin Holmberg, Fredrik Winquist, Ingemar Lundström, Julian W. Gardner, and Evor L. Hines, Sensors and Actuators B 26-27 (1995) 246-249.

Load the Neural Networks package and the data.

There is one data set for estimation, xs and ys, and one for validation of the classifiers, xv and yv.

Check the dimensions of the data.

There are 48 data samples available for estimation. Each paper sample is characterized by 15 x values from the 15 sensors in the electronic nose sensor array.

Five different types of cardboard paper and plain air are measured, making six possible output classes. The correct class of each data sample is stored in y, with a 1 in the appropriate column indicating the class of the sample.

Check the class of the 27th sample of validation data.

The 27th sample belongs to class 3.

It is always a good idea to check how many data samples there are from each class.

Look at the distribution of estimation data over the classes.

Look at the distribution of validation data over the classes.