Feranmii/

XGboost_CreditCard_Fraud

XGBoost Model for Fraud Detection This repository contains an XGBoost model trained to detect fraudulent transactions based on transaction data. The model uses several engineered features that are designed to capture various aspects of the transactions, including anonymized features, time-based features, and behavioral features.

Features Used in the Model The model is trained on the following features, which include both anonymized PCA-like features (V1-V28) as well as some time and behavioral features:

V1-V28 Features These are anonymized features generated using Principal Component Analysis (PCA) on the raw transaction data. They are designed to capture latent patterns in the data that are not directly observable in the original features. The anonymization process ensures that no personal or sensitive information is exposed.

V1-V28: These are features generated from the raw transaction data (like transaction amount, frequency, etc.), and each of them represents a component extracted via PCA or similar methods to anonymize the data. These features help the model learn complex patterns in the data without exposing the original sensitive attributes. Other Features in the Dataset DOW (Day of Week): The day of the week on which the transaction occurred. Values range from 0 (Monday) to 6 (Sunday). Is_Weekend: A binary feature indicating whether the transaction occurred on the weekend (1) or on a weekday (0). Max_Days_Between: The maximum number of days between this transaction and the previous transaction made by the same user. Monthly_Txns_lag_30: The cumulative number of transactions made by the user in the last 30 days. Weekly_Txns_lag_30: The cumulative number of transactions made by the user in the last 7 days. Class_lag_30: A placeholder feature that represents the fraud class from 30 days ago. Since fraud labels are not available historically, it is set to 0. Model Evaluation The model has been evaluated on a dataset of 10,236 transactions, with a significant class imbalance, as fraudulent transactions (label 1) make up a small portion of the data.

Confusion Matrix The confusion matrix summarizes the performance of the classification model:

lua Copy Edit [[6487 13] [ 6 3730]] True Negatives (TN): 6487 False Positives (FP): 13 False Negatives (FN): 6 True Positives (TP): 3730 Classification Report The classification report provides a detailed breakdown of the model's precision, recall, and F1-score for each class (fraudulent and non-fraudulent):

markdown Copy Edit precision recall f1-score support

       0     0.9991    0.9980    0.9985      6500
       1     0.9965    0.9984    0.9975      3736

accuracy                         0.9981     10236

macro avg 0.9978 0.9982 0.9980 10236 weighted avg 0.9981 0.9981 0.9981 10236 Class 0 (Non-Fraudulent Transactions):

Precision: 0.9991 Recall: 0.9980 F1-score: 0.9985 Class 1 (Fraudulent Transactions):

Precision: 0.9965 Recall: 0.9984 F1-score: 0.9975 ROC AUC Score The model achieved a ROC AUC Score of 1.0, indicating that it has perfect discriminatory power between fraudulent and non-fraudulent transactions. This means the model is able to perfectly distinguish between both classes across all possible thresholds.

Model Performance Summary Accuracy: 99.81% Precision (Class 0): 99.91% Recall (Class 0): 99.80% Precision (Class 1): 99.65% Recall (Class 1): 99.84% F1-Score (Class 1): 99.75% ROC AUC Score: 1.0 The model performs excellently in distinguishing fraudulent transactions, with an extremely high accuracy and very low false positive and false negative rates.