Assignment 4: Machine learning problem for credit card fraud

Assignment 4:

1 Question 1 (30 pts)
: In this question we are going to design a machine learning problem for credit card fraud detection. You have to predict, based on certain attributes, whether a transaction is fraudulent or
legitimate. Answer the following questions.
• 1 a. Is this a classification problem or a regression problem? (2 pts)
• 1 b. Describe at least 4 discrete features of each transaction that is going to help in our
problem. (8 pts)
• 1 c. Describe at least 3 continuous features of each transaction that is going to help in our
problem. (6 pts)
• 1 d. If we were to use an ML algorithm that only takes discrete features as input, what can
we do to the continuous features you described in 1 (c) so that they can be used by the ML
algorithm? (4 pts)
• 1 e. How would you know if your model suffers from overfitting? If you are using decision
tree in your problem, how would you deal with overfitting? (4 pts)
• 1 f. In Table 1, observe the data and use your own intuition to draw a decision tree that can
classify the data into Class 1 and Class 2. (6 pts)
2 Question 2 (40 pts)
:
1. The input dataset is : ballons.csv
2. The output file is: output ballons.txt
3. Starter code: q2 decision tree.py
You can find the dataset and starter code in CMSC 471 Assignment 4.zip on blackboard.
You have to implement the code to choose the best feature to split the dataset in a decision tree.
Please use entropy and information gain as the metrics to choose a feature. You will find some
starter code in q2 decision tree.py. Implement the functions that are marked with ‘#TODO’.
Note: Your python code should NOT take any input parameter. It should print the output in
a file ‘output.txt’. If your python code fails to run without any input parameters, you will NOT
be graded for this question.
Table 1: Table with variables A,B,C and Output Label
A B C Output
1 1 1 Class 1
1 1 0 Class 1
0 0 1 Class 2
1 0 0 Class 2
3 Question 3 (30 pts)
1. The input training dataset is : titanic train.csv
2. The test dataset is: titanic test.csv
3. The output file is: output titanic.txt
For this question, consider the Titanic dataset. You have to predict whether a person will
survive based on the features of that person.
You may use sklearn’s models for training. Please use titanic train.csv. for training and
report your scores on titanic test.csv.
If you want to reduce the complexity, you may drop the columns ‘Name’, ‘Ticket’, and ‘Cabin’.
Please note that the feature ‘Passenger Id’ is an identifier field. For fields with missing values,
use np.mean or np.median to impute the missing values. For example if the field age has 5 values
10,10,20,20, [missing], use 15 to impute the missing value.
Train the following ML models (at least) on the training dataset
• Random Forest (10 pts)
• Logistic Regression (10 pts)
• SVM (10 pts)
For each of these models, submit your classification report (use sklearn.metrics.classification
report) on the test dataset (test.csv). Write your classification report to the ‘output titanic.txt’
file.
You can find the datasets in CMSC 471 Assignment 4.zip on blackboard.
Note: Your python code should NOT take any input parameter. It should print the output in
a file ‘output titanic.txt’. If your python code fails to run without any input parameters, you
will NOT be graded for this question. You have to ensure that the file that you read (train or test)
is in the same directory as your code.
Extra Credits (20 pts): Explore other models in the sklearn library, run them and report
any other model that performs better than the best performing model that you encountered in
Question 3. Report the classification report on the test dataset for this model.
2