leilamr / fisheriris-mlp

Iris flower classification with MLP using MATLAB.

Dataset Iris plant

Iris flower classification with MLP using MATLAB.

Atribute informations

1. sepal length in cm
2. sepal width in cm
3. petal length in cm
4. petal width in cm
5. class: Iris Setosa, Iris Versicolour and Iris Virginica.

Class coding

• Setosa = [1 0 0 ]
• Versicolor = [0 1 0]
• Virginica = [0 0 1]

Network settings:

Train = 70%, Validation = 15% and Testing = 15%
Number hidden of nodes = 4 
Epochs = 1000
Trainng Function = trainlm
Transfer Function (layer 1) = tansig
Trasnfer Function (layer 2) = purelin

Accuracy = 99.3%

Cross-validation: k-fold

In fisherIris_mpl_kfold.m the dataset was divided into 10 folds. Each k-folds has size 15x5.

The best configuration obtained from the network with the cross validation technique was:

Number hidden of nodes = 4 
Epochs = 1000
Trainng Function = trainlm
Transfer Function (layer 1) = tansig
Trasnfer Function (layer 2) = purelin

-- Average accuracy = 94.667%

References

https://la.mathworks.com/help/deeplearning/gs/classify-patterns-with-a-neural-network.html https://la.mathworks.com/help/deeplearning/ref/plotconfusion.html https://la.mathworks.com/help/deeplearning/ref/patternnet.html