Many thousands of audits of taxpayers' tax returns might be performed each year. The outcome

of an audit may be productive, in which case an adjustment to the information supplied was

required, usually resulting in a change to the amount of tax that the taxpayer is liable to pay $($an

increase or a decrease$).$ An unproductive audit is one for which no adjustment was required after

reviewing the taxpayer's affairs. The audit dataset attempts to simulate this scenario. It is

supplied as a CSV file. The dataset consists of $2,000$ fictional taxpayers who have been audited

for tax compliance. For each case, an outcome of the audit is recorded in $“$TARGET$\_$Adjusted$”,$

i$.$e$.,$ whether the financial claims had to be adjusted or not. The actual dollar amount of any

adjustment that resulted is also recorded in $“$RISK$\_$Adjustment$”($noting that adjustments can go

in either direction$).$ The audit dataset contains $12$ variables, with the first variable being a unique

client $“$ID$”,$ followed by $9$ input variables from $“$Age$”$ to $“$Hours$”.$

Please NOTE: $1)“$RISK$\_$Adjustment$”$ is only used for generating the profit curve. Do not use it as

an input variable when developing a classifier; $2)“$TARGET$\_$Adjusted$”$ is the target variable with

$1$ for Yes as the positive class and $0$ for No as the negative class.

You are asked to use this dataset accomplish the following tasks:

$1)$ Load the dataset into Rattle and impute the missing values of $“$Employment$”$ and

$“$Occupation$”.$ Show in your report a screenshot of your variables $($after imputation$)$ under

the Data tab in Rattle.

$2)$ Choose a seed $($note: show it in your report$).$

$3)$ Find the BEST SVM model to predict the target variable, $“$TARGET$\_$Adjusted$”,$ by tuning

the parameters using the validation set $($of the default partition of $“70/15/15”)$ and the

area under the ROC curve $($AUC$)$ as your evaluation criterion $($noting that: the higher AUC

score, the better performance$).$ Show your parameters and AUC scores in a table $($no need

to show the ROC plots$).$ 

$4)$ Then set the partition to $“85/0/15”,$ and rebuild your SVM model using the parameters

that give you the highest AUC score in Task $3.$ Show the ROC plot of this model on the

testing dataset.

$5)$ Finally, use the model in Task $4$ and the testing dataset to generate a profit curve. Assume

that the cost for auditing a return is $$3000,$ and that the $“$RISK$\_$Adjustment$”$ column

contains the amount recovered from a successful audit, which is used to estimate the

benefit. Based on your profit curve, give the IRS a specific recommendation for what

percentage of returns should be audited