SharmaineMangombe/Tunisian-Fraud-Detection-Challenge

# Tunisian Fraud Detection Challenge ## Project Overview This project focuses on detecting fraudulent transactions in a Tunisian financial dataset. Using machine learning techniques, the goal is to identify patterns and anomalies in transaction data to minimize fraud risk and improve financial security. ## Dataset The dataset contains transaction records with features such as: - Transaction ID - Customer ID - Transaction amount - Date and time - Other anonymized transaction-related features ## Objective Develop a predictive model to classify transactions as **fraudulent** or **non-fraudulent**, aiming to: - Maximize detection of fraudulent transactions - Minimize false positives ## Approach ### 1. Data Cleaning & Preprocessing - Handled missing values and inconsistencies - Normalized and scaled numerical features - Encoded categorical variables ### 2. Exploratory Data Analysis (EDA) - Analyzed transaction patterns - Visualized feature distributions and correlations - Identified key indicators of fraud ### 3. Modeling - Tested multiple machine learning algorithms: - Logistic Regression - Random Forest Classifier - Gradient Boosting (XGBoost) - Tuned hyperparameters to optimize performance ### 4. Evaluation - Used metrics: Accuracy, Precision, Recall, F1-Score, ROC-AUC - Focused on maximizing fraud detection (Recall) while minimizing false positives ### 5. Deployment & Insights - Generated predictions for unseen test data - Identified high-risk transactions for further investigation ## Tools & Technologies - Python: `pandas`, `NumPy`, `scikit-learn`, `XGBoost`, `matplotlib`, `seaborn` - Jupyter Notebook - GitHub for version control ## Results - Achieved **[insert your metric here, e.g., F1-score: 0.85]** on validation data - Key insights: High transaction amounts during unusual hours were strong indicators of fraud