Regression Task $($predicting continuous numerical variables$)(50\%-100$ marks$)$

For this task, you will find a dataset and predict a specific value using $3$ different types of Machine

learning algorithms $($one must be using neural networks$)$ and write a small report. Your first sub$-$task

is to find a dataset suitable for regression $($tabular data with mainly numerical values$).$ For regression

models, unless you are using a neural network $($even then, it can vary$),$ you will not need to

normalise the data unless your research says otherwise. From your dataset, print out a scatter

matrix of all numerical data. You should also aim to visualise any string data too. You will need to

analyse the string data, group them so you can generate numerical data from it$,$ then create some

graphs. Even though you may have string data in the dataset, it may not be useful for regression so

you will not be penalised for this task if you do not use that data type. You then need to pick $3$

algorithms; these can be any of your choice, but you will need to explain how they work in your

report. Once you have fit the model and made some appropriate predictions, state the accuracy of

the model after using the test data, and evaluate its usefulness for your dataset. In the report, you

must explain in detail as to why the model has performed in such way and suggest ways of

improving the model.

Create a regression model using algorithm $1$ state which algorithm and submit working source

code $(10$ marks$)$ Model can be either Support Vector Regressor, Decision Tree Regressor, Random

Forest Regressor or Multiple Linear Regression $($you are not limited to these, but these will be taught

on the module$).$

Create a regression model using algorithm $2$ state which algorithm and submit working source

code $(10$ marks$)$ Model can be either Support Vector Regressor, Decision Tree Regressor, Random

Forest Regressor or Multiple Linear Regression $($you are not limited to these, but these will be taught

on the module$).$

Create a model using Neural Networks $$ submit working source code $(20$ marks$)$

Report $(60$ marks$)$

$1.$ Showing and comparing accuracies of each model. $(10$ marks$)$

$2.$ Explanation of how your two models work $($include references$)(25$ marks$)$

$3.$ Suggest on different methods to improve your models. E$.$g$.$ Remove$/$add data? If so$,$ which

features would you build on$?$ Think about visualising the data first and seeing which features

do not help with the prediction. For NN models, change the number of neurons, layers etc?

$(25$ marks$)$