Use the Mathematica toolbox (Wolfram) for Neural networks with feedforward neural networks and prepare the classification of the iris data set. The set is available at [login to view URL] or it is accessible in Mathematica too via e.g. the following commands:
Try to train the iris dataset within several hidden nodes and different transfer functions, in a hidden layer and an output layer. (e.g. Neuron-> Sigmoid, OutputNonlinearity->Tanh)
possibilities: Sigmoid, Tanh, SaturatedLinear
Make a conclusion based on the results.
a) training data for training,
b) testing data for testing....
c) you can check the training and testing error and make a conclusion if your neural network suffers with overfitting.
d) make a graph of the global error
e) visualize the distribution via classes (NetPlot with DataFormat -> BarChart)
f) prepare the confusion matrix - true positive, false positive... items.
The comments can be written directly in Mathematica notebook including the conclusion.