ReraA/Fraud-Detection-Nigeria

# Machine Learning Model for Detecting Fraudulent Transactions in Nigeria's Digital Banking Sector Seminar research project. Trains and compares three supervised machine learning models (Logistic Regression, Random Forest, XGBoost) for fraud detection on the PaySim synthetic mobile money dataset, following the methodology described in Chapter Three of the accompanying paper. ## 1. Get the dataset This repository does not include the dataset (it is ~470MB and hosted on Kaggle under its own license). To get it: 1. Create a free Kaggle account if you don't have one: kaggle.com 2. Go to: kaggle.com 3. Click **Download** and unzip it. 4. Place the CSV file in this repo's `data/` folder and rename it to: `data/paysim.csv` Alternatively, using the Kaggle CLI (`pip install kaggle`, after setting up your `kaggle.json` API token as described at github.com): ```bash kaggle datasets download -d ealaxi/paysim1 -p data --unzip mv data/PS_20174392719_1491204439457_log.csv data/paysim.csv ``` ## 2. Set up the environment ```bash python3 -m venv venv source venv/bin/activate # on Windows: venv\Scripts\activate pip install -r requirements.txt ``` ## 3. Run the pipeline ```bash python src/train_evaluate.py ``` This will: - load and clean `data/paysim.csv` - engineer additional features - split the data (stratified) into train/test sets - apply SMOTE to the training set only - train Logistic Regression, Random Forest, and XGBoost with grid-searched hyperparameters - evaluate all three on the untouched test set using precision, recall, F1-score, and ROC-AUC - save a comparison table, ROC curves, and confusion matrices to `outputs/` Expect this to take anywhere from a few minutes to a while longer depending on your machine, the dataset has about 6.3 million rows. ## 4. What you'll get in `outputs/` - `model_comparison.csv` — the precision/recall/F1/ROC-AUC table for all three models, this is th …