Dataset 1: 2-dimensional artificial data:
(a) Linearly separable data set for static pattern classification
(b) Nonlinearly separable data set for static pattern classification
Dataset 2: Image data set for static pattern classification
Classifiers to be built for Dataset 1(a):
- Perceptron for every pair of classes
- Multilayer feedforward neural network (MLFFNN) with a single hidden layer for all classes
- Linear SVM classifier for every pair of classes
Classifiers to be built for Dataset 1(b) :
- MLFFNN with two hidden layers
- Nonlinear SVM using one-against-the-rest approach : (a) Polynomial kernel, (b) Gaussian
kernel
Classifiers to be built for Dataset 2:
- MLFFNN with two hidden layers
- Gaussian kernel based SVM using one-against-the-rest approach
Use the cross-validation method to choose the best values of hyperparameters.
Report should include the following for each classifier and for each dataset:
- Table of classification accuracies of the model on training data and validation data for
different values of hyperparameter
- Classification accuracy of the best configuration of the model on test data
- Confusion matrix for the best configuration of the model, on training data and test data
- Decision region plots for Datasets 1(a) and 1(b). Superpose the training data on the
decision region plot. For SVM model, mark the support vectors.
- For the best configuration of MLFFNN classifier on Dataset 1(b), plot the surfaces of the
outputs of hidden layer nodes and output layer nodes after the following epochs: 1, 5, 20, 100, and after convergence.
Report should also include your observations about the performance for each classifier, and for each dataset.
