Scribd Logo
Upload

Welcome to Scribd!

  • 4 views

    Task 5

    Uploaded by

    azurebirble12
    This document discusses classifying credit card transactions as fraudulent or not fraudulent using gradient boosting algorithms. It involves preprocessing a credit card transaction dataset, training classifiers like XGBoost and logistic regression on the data, and evaluating the classifiers' performance using metrics like accuracy, precision, recall and F1 score.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd

    Task 5

    Uploaded by

    azurebirble12
    4 views2 pages
    This document discusses classifying credit card transactions as fraudulent or not fraudulent using gradient boosting algorithms. It involves preprocessing a credit card transaction dataset, training classifiers like XGBoost and logistic regression on the data, and evaluating the classifiers' performance using metrics like accuracy, precision, recall and F1 score.

    Original Title

    Task5

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    This document discusses classifying credit card transactions as fraudulent or not fraudulent using gradient boosting algorithms. It involves preprocessing a credit card transaction dataset, training classifiers like XGBoost and logistic regression on the data, and evaluating the classifiers' performance using metrics like accuracy, precision, recall and F1 score.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    4 views2 pages

    Task 5

    Uploaded by

    azurebirble12
    This document discusses classifying credit card transactions as fraudulent or not fraudulent using gradient boosting algorithms. It involves preprocessing a credit card transaction dataset, training classifiers like XGBoost and logistic regression on the data, and evaluating the classifiers' performance using metrics like accuracy, precision, recall and F1 score.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    of 2

    Section 5: Credit Score Classification with Gradient Boosting (25 Points)

    Load the dataset creditcard_fraud_subsample.csv from STiNE. The dataset contains trans-
    actions made by credit cards in two days in September 2013 by european cardholders. It
    contains only numerical input variables which are the result of some transformation (due to
    confidentiality issues there is not more background on the original features available). The
    only features which have not been transformed are Time and Amount, which display the seconds
    elapsed between each transaction and the first transaction in the dataset and the transaction
    amount. The Feature Class is the response variable and it takes value 1 in case of fraud and
    0 otherwise.

    1. The first task is to preprocess the dataset. Remove the feature ‘Time’ from the dataset
    and standardize the feature ‘Amount’ by substracting the mean and scaling to unit
    variance.
    2. Split the data into a training and test set to be able to test the out of sample performance
    (use 40 percent of the data for the test set).
    3. Do some research on the classification method called “Gradient Boosting”, for example
    using the book “Elements of Statistical Learning”. Explain the main idea and describe
    how the algorithm proceeds.
    4. We will use a version of gradient boosting that is called extreme gradient boosting. A pop-
    ular implementation is XGBoost, see https://xgboost.readthedocs.io/en/stable/python/
    python_api.html#xgboost.XGBClassifier. Hint: You have to install the XGBoost pack-
    age using pip install xgboost, more info available at https://xgboost.readthedocs.
    io/en/stable/install.html#python. If you have problems with installing XGBoost, you
    can also use scikit-learn’s GradientBoostingClassifier, see https://scikit-learn.
    org/stable/modules/generated/sklearn.ensemble.GradientBoostingClassifier.html.
    5. Use the XGBoostClassifier to predict the value of Class based on the available fea-
    tures. Compute the accuracy. Compare the predictive performance to that of a logistic
    regression. Also compare the accuracy with a naive classifier that simply ‘predicts’ “no
    fraud” in every possible case. Why is the accuracy of the second classifier still very close
    to one?
    6. To be able to evaluate the performance we define the confusion matrix (binary classifi-
    cation, 1 =̂ positive)

    𝐶1,1 𝐶1,0 true positives false negatives


    𝐶 ∶= ( )=( ),
    𝐶0,1 𝐶0,0 false positives true negatives

    where 𝐶𝑖,𝑗 contains the number of observations of the test sample, which are in class 𝑖
    and classified as class 𝑗. Write a function which takes two vectors (where each entry is
    either zero or one) as input and calculates the confusion matrix. Use your function to
    calculate the confusion matrix on the test set for both classifiers.

    7
    7. We define
    𝐶1,1
    precision ∶=
    𝐶1,1 + 𝐶0,1
    and
    𝐶1,1
    recall ∶= .
    𝐶1,1 + 𝐶1,0
    What do precision and recall measure? Write functions which take two vectors as input
    and calculate precision and recall. Use your functions to evaluate your classifier (here
    nan is considered a valid result if you divide by zero).
    8. Finally, we combine precision and recall to the 𝐹1 -score

    precision ⋅ recall
    𝐹1 ∶= 2 .
    precision + recall
    Write a function which takes two vectors as input and calculates the 𝐹1 -score. Try
    to train new classifiers (e.g. Random Forest, SVM, Logistic Regression) with different
    tuning parameters and maximize the 𝐹1 -score on the test sample.
    9. (Bonus): Explain the role of the parameters n_estimators and learning_rate in
    XGBoost. Set the learning_rate to a value of 0.1. We want to find the optimal
    value of n_estimators, i.e., the one that gives the highest 𝐹1 -score. Provide your own
    implementation of cross-validation from scratch in order to find the best possible value
    for n_estimators based on the cross-validated 𝐹1 score.

    You might also like