For this coursework you will implement in MATLAB a Multilayer Neural Network for predicting the quality of red wines based on physicochemical tests. The quality is a value between 1 and 10; therefore you will treat this as a regression problem – i.e. trying to predict a value between 1 and 10 that is as near as possible to the correct value. You are given a dataset (winequality-red.train.txt and winequality-red.test.txt) consisting of 1000 training and 599 test examples and your aim is to train a network that predicts as closely as possible the values of the test examples. To do this you should try many different settings: different number of hidden units, size of validation examples, normalization type, initial weights, and optionally activation and/or training functions.

Deliverables:

Your code in electronic form – a zip file containing all files.

A report describing your experiments and your findings (including all the different settings you tried, their results and any conclusions you can make based on your experiments).