Exercise 1 : Building a Feed-Forward Neural Network Exercise 1.1 : Data Preprocessing
¡ñ As the classes are categorical, use one-hot encoding to represent the set of classes. You will find this useful when developing the output layer of the neural network.
¡ñ Normalize each field of the input data using the min-max normalization technique.
Exercise 1.2 : Training and Testing the Neural Network
Copyright By PowCoder代写 加微信 powcoder
Design a 4-layer artificial neural network, specifically a feed-forward multi-layer perceptron (using the sigmoid activation function), to classify the type of ‘Dry Bean’ given the other attributes in the data set, similar to the one mentioned in the paper above. Please note that this is a multi-class classification problem so select the right number of nodes accordingly for the output layer.
For training and testing the model, split the data into training and testing set by 90:10 and use the training set for training the model and the test set to evaluate the model performance.
Consider the following hyperparameters while developing your model :
¡ñ Number of nodes in each hidden layer should be (12, 3)
¡ñ Learning rate should be 0.3
¡ñ Number of epochs should be 500
¡ñ The sigmoid function should be used as the activation function in each layer
¡ñ Stochastic Gradient Descent should be used to minimize the error rate
Requirements once the model has been trained :
¡ñ A confusion matrix for each class, specifying the true positive, true negative, false positive,
and false negative cases for each category in the class
¡ñ The accuracy and mean squared error (MSE) of the model
¡ñ The precision and recall for each label in the class
¡ñ Splitting of the dataset should be done after the data preprocessing step.
¡ñ The mean squared error (MSE) values obtained should be positive.
¡ñ All performance metrics should be calculated with respect to the testing dataset.
程序代写 CS代考 加微信: powcoder QQ: 1823890830 Email: powcoder@163.com