Advanced Neural Networks in R - A Practical Approach

A quick view in the world of (maybe) the most popular neural networks - multilayer perceptrons. What they are and how their structure looks like.

Let's learn more about summation and activation, the more important processes that take place in a neural network.

The learning process of a multilayer perceptron, in layman's terms.

How to measure the predictive performance of MLPs - a few simple indicators.

Here's another powerful tool for assessing the prediction performance of a neural network when the response variable is categorical: the ROC curve. Let's see what it is and how to interpret it.

How to set the parameters of a multilayer perceptron when the response variable is qualitative. And how to prepare your data properly.

How to use our MLP to make predictions in the test set.

Learn the code for creating ROC curves in R and computing the area under curve (it's simple).

Let's try to improve the prediction accuracy of our network by changing the number of hidden nodes in the hidden layer.

Finally, let's validate our predictive model using the k-fold validation technique. Learn a few commands in the caret package that make k-fold validation very easy.

How to set the parameters of MLP when the response variable is numeric.

How to make predictions and assess the predictive accuracy of our MLP in the test set.

Let's see if we can improve the predictive performance of our network by modifying the number of nodes in the hidden layer.

Validating our MLP using the k-fold technique.

A short theoretical introduction: the structure of a probabilistic neural network and how it works.

How to prepare your data properly so you can use it as an input for a PNN.

How to create a probabilistic neural network in R, step by step.

How to make predictions in the training set using our new PNN.

Improve the prediction accuracy of our PNN by changing the value of the smoothing parameter sigma.

After identifying an optimal sigma, we validate our PNN using the k-fold cross-validation technique.

A short theoretical introduction: the structure of a generalized regresssion neural network and how it works.

How to arrange your data so it can be used for GRNNs.

All the commands you need to build and train a GRNN in the R program.

Let's evaluate the prediction accuracy of our new GRNN in the training set.

Identify the smoothing parameter value that maximizes the prediction accuracy of our network.

A short theoretical introduction: the structure of a recurrent neural network and how it works.

In this lecture we introduce two specific metric for measuring the predictive performance in time series: mean absolute standard error and mean absolute percentage error.

How to prepare your data for creating an Elman recurring neural network. We are going to use a package called quantmod to create lagged variables.

How to build and train an Elman network in R.

How to do times series forecasting with our new Elman network, using specific indicators.

We'll try to improve the prediction performance of our Elman network by adding more lagged variables as predictors.

Assessing the predictive accuracy in the training set after adding more variables to the model.

Prepare your data so you can use it as an input in a Jordan recurrent neural network.

Time to build and train our Jordan recurrent network.

Let's perform time series forecasting with our Jordan network.

In this lecture you have a detailed description of all variables in the data sets used for the practical exercises.

You can find them in the PDF document attached to this lecture.

Download the data sets and the source code for all lectures.