Description

In each assignment you will be able to earn up to 5 marks based on your engagement on the moodle Discussion forums from the topics associated with the given assignment. See the Assignment assessment criteria document for more details.

Question 2

The Toy data-set is a simulated data-set that contains a response variable, Y , along with nine predictor variables. The following tasks require you to use the neuralnet package in R. Use a 75:25 train:test split for parts (a) and (b). Use a set.seed value of 430 for all questions.

• Train a neural network with only one hidden unit to predict the response variable using all predictor variables. Produce a graph of your trained neural network. Provide an overview of your analysis and results, including the correlation coefficient of actual and predicted Y in the data-set.
• Train 7 neural networks, each time adding one extra hidden node. Produce a summary graph of the correlation coefficient of each network vs the number of hidden units. Provide an overview of your analysis and results.
• Train 4 neural networks with differing training and testing splits as follows: [20% training; 80% testing], [40% training; 60% testing],[60% training; 40% testing], [80% training; 20% testing]. Produce a summary graph of the correlation coefficient of each network vs the percentage of training set data used. Provide an overview of your analysis and results.

Hint: for parts (b) and (c) use a for loop in your code to automate the process