Consider the dataset $($in csv format$)$ that describes the loan approval history for past applicants.

Important Instructions:

$1.$ Answer all tasks in a single Python notebook file.

$2.$ You will need to upload your Python notebook $($in ipynb format$)$ with all outputs embedded and the corresponding pdf file.

$3.$ Clearly create text cells of sub$-$headings in the single ipynb code file for each task.

$4.$ You can use any python based libraries of your choice to solve the tasks unless otherwise stated.

Task $1$: Data Visualization $($DV$)$: Load the dataset. Illustrate, via suitable visualizations, distributions of gender and ethnicity with respect to the approval decisions. $[1$ mark$]$

Task $2$: Fairness Metric $($FM$)$: Assume gender as sensitive attribute, write your "own" functions to compute group$-$level fairness metrics $($absolute difference$)$: demographic parity, equality of odds, and equality of opportunity. Use logistic regression model to get predicted 'approved' values. $[2$ marks$]$

Task $3$: Fairness Metric $($FM$)$: Assume ethnicity as sensitive attribute, write your "own" functions to compute group$-$level fairness metrics $($max $-$min difference$)$: demographic parity, equality of odds, and equality of opportunity. Use logistic regression model to get predicted 'approved' value. $[$max $-$ min difference is the absolute difference between highest and lowest probabilities among ethnicity groups$][3$ marks$]$

Task $4$: Prejudice Removal Regularizer $($PRR$)$: Incorporate prejudice removal regularizer with $4$ appropriately chosen regularization constants and report the accuracy and group$-$level fairness metrics using $5$ fold cross validation for all chosen regularization constants. Use any existing library function to implement PRR$.$ Assume gender as sensitive attribute. $[4$ marks$]$