Credit Card Transactions Fraud Detection

Dataset on kaggle

Contents

Data Understanding

• Features
• Null values detection
• Duplicated values detection

Data Cleaning For EDA

• Column removal
• Discretization
• Creating new features
• Feature extraction

Feature Engineering

• Datetime feature extraction
• Credit card feature extraction

Exploratory Data Analysis

• Univariate Analysis

  • Target
  • Categorical features
  • Numerical features

• Bivariate Analysis

  • Target analysis
  • Amount of activity analysis
  • Time analysis

Correlation and Association Analysis

• Correlation matrix
• Association matrix

Data Preprocessing

• Column removal

• Log transform

• Categorical encoding

  • Binary encoding
  • Weight of evidence encoding
  • Ordinal encoding

• Train-test split

Imbalanced Learning

Target is imbalanced

• 

Methods performed

• No changes
• Random under sampling
• Random over sampling
• SMOTE-Tomek links
• Class weights

Feature Importance

Modeling

1. Random Forest Classifier
2. Logistic Regression Classifier
3. Naive Bayes
4. Decision Tree Classifier
5. Support Vector Machine (SVM) Classifier
6. K-nearest neighbor (KNN) Classifier

Evaluation

• Confusion matrix
• AUC curve
• Classification metrics
• Decision boundary

Results on random forest classifier for test data

• 

• 

Model Selection

Results on all models for test data