https://github.com/absarraashid3/fraudscope-

FraudScope is a complete machine learning pipeline for detecting fraudulent financial transactions using only classical ML techniques and core concepts. Designed as a hands-on, transparent system for fraud analytics, it guides users through every stage—from raw data to model evaluation—using only scikit-learn and core Python libraries.
https://github.com/absarraashid3/fraudscope-

Last synced: about 1 year ago
JSON representation

FraudScope is a complete machine learning pipeline for detecting fraudulent financial transactions using only classical ML techniques and core concepts. Designed as a hands-on, transparent system for fraud analytics, it guides users through every stage—from raw data to model evaluation—using only scikit-learn and core Python libraries.

• Host: GitHub
• URL: https://github.com/absarraashid3/fraudscope-
• Owner: AbsarRaashid3
• Created: 2025-06-24T20:03:38.000Z (about 1 year ago)
• Default Branch: main
• Last Pushed: 2025-06-24T20:06:30.000Z (about 1 year ago)
• Last Synced: 2025-06-24T21:23:44.229Z (about 1 year ago)
• Language: Jupyter Notebook
• Size: 1.21 MB
• Stars: 0
• Watchers: 0
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# FraudScope: Transaction Fraud Detection & Analysis

FraudScope is a complete machine learning pipeline for detecting fraudulent financial transactions using only classical ML techniques and core concepts. Designed as a hands-on, transparent system for fraud analytics, it guides users through every stage—from raw data to model evaluation—using only scikit-learn and core Python libraries.

# Dataset & Problem Statement

Dataset Source: MoMTSim_20240722202413_1000_dataset.csv (synthetic transactional data)

Objective: Build a classifier to detect fraudulent transactions (isFraud)

Features: Includes transaction metadata such as:

amount, oldBalInitiator, newBalInitiator

transactionType, origin, destination, etc.

# Architecture Overview

A fully modular pipeline:

Raw CSV Data

   ↓

Exploratory Data Analysis (EDA)

   ↓

Handling Missing Values & Duplicates

   ↓

Outlier Detection (IQR & Z-score methods)

   ↓

Encoding & Feature Engineering

   ↓

Feature Scaling (MinMaxScaler, StandardScaler)

   ↓

Feature Selection (Correlation Matrix Filtering)

   ↓

Train/Test Split (80/20, stratified on 'isFraud')

   ↓

Model Training & Hyperparameter Tuning:

   - K-Nearest Neighbors (varied k, distance metrics)

   - Support Vector Machine (kernel, C tuning)

   - Decision Tree (depth, min_samples_split)

   - Logistic Regression (L1 & L2 regularization)

   ↓

Evaluation:

   - Accuracy, Precision, Recall, F1-score

   - Confusion Matrix & Classification Report

   - Training & Inference Time Comparison

     

# Setup & Requirements

Install dependencies with:

bash

Copy

Edit

pip install -r requirements.txt

Required libraries:

pandas, numpy

matplotlib, seaborn

scikit-learn

 # Notebook Walkthrough

fraudscope.ipynb:

EDA: Value counts, missing data, class imbalance

Visualizations: Boxplots, heatmaps, pairplots

Outlier Handling: IQR and Z-score thresholds

Feature Engineering: Encoding transaction types, calculating balance deltas

Scaling: Compared MinMax and Z-score scaling effects

Model Training: Manual loop for each model with grid-tuned hyperparameters

Evaluation: Comparative analysis of model metrics and runtime

# Results & Model Comparisons

Model	F1-Score	Accuracy	Inference Time

Logistic Regression	0.92	0.94	⚡ Fast

KNN (k=5, Euclidean)	0.88	0.91	🐢 Slow

Decision Tree	0.93	0.95	⚡ Fast

SVM (RBF)	0.94	0.96	⚖️ Moderate

Best performer: SVM (RBF kernel) in terms of F1 and overall accuracy

Most efficient: Logistic Regression (fastest inference)

# Usage & Reproducibility

To run the notebook:

Place the dataset (MoMTSim_20240722202413_1000_dataset.csv) in the root directory.

Open and run fraudscope_analysis.ipynb in Jupyter or Google Colab.

Follow the cell-by-cell analysis and observe output metrics.

# Contributing

Feel free to fork this repo and submit a pull request if you'd like to:

Add new models (e.g., ensemble methods)

Improve feature engineering

Visualize confusion matrices interactively

# License & Academic Integrity

This project is for educational and demonstrative purposes only. Do not reuse directly in academic submissions unless explicitly permitted.

# Developed by

Absar Raashid

GitHub Repository

Open for feedback and collaboration.