Assignment-10---Supervised-Machine-Learning-Model

Mini Project: Supervised Machine Learning – Fraud Detection

Project Overview

This project applies supervised machine learning techniques to detect fraudulent financial transactions using the fraud.csv dataset. It involves exploring, pre-processing, modelling, and evaluating the dataset to build a system capable of identifying fraudulent behaviour with high accuracy and reliability.

---

🧾 Project Requirements & Description

This project follows the key requirements of the assignment brief:

1. Data Exploration

   • Analysed structure and distribution of fields in fraud.csv
   • Identified zero and missing values
   • Detected the need for scaling and encoding due to imbalance and categorical data

2. Data Preparation

   • Filled zero values with medians where appropriate
   • Scaled data using RobustScaler to manage outliers
   • Encoded categorical features using LabelEncoder
   • Sampled data to address class imbalance for model fairness

3. Model Development

   • Trained multiple supervised ML models:
     • Logistic Regression (regular)
     • Support Vector Machines (Linear, RBF, Sigmoid)(undersampled)
     • Random Forest (regular and undersampled)
     • Neural Network (undersampled)

4. Model Evaluation

   • Used accuracy, precision, recall, and F1-score
   • Plotted ROC curves for all models
   • Analysed trade-offs between false positives and false negatives

---

Dataset Used

• Name: fraud.csv
• Description: A simulation of mobile money transactions labelled as fraud or non-fraud.
• Source: Kaggle – PaySim Fraud Detection Dataset

---

Models & Results Summary

Model	Precision	Recall	F1-Score	Accuracy
Logistic Regression	Low	High	Low	97%
SVM (Linear/RBF)	~72%	~90%	Strong	98%
Random Forest	13%	98%	Moderate	99%
RF (Undersampled)	99%	97%	Best	100%
Neural Network	94%	88%	High	98%

• ✅ Best Model: Random Forest (Undersampled)
• 🚫 Worst Kernel: SVM Sigmoid — struggled with precision and overall accuracy
• 📉 Accuracy alone was misleading due to class imbalance; precision/recall were prioritized

---

Next Steps

• Fine-tune top models using GridSearchCV
• Add ‘is_off_hours’ feature (e.g., 0–6 AM) to allow model to flag time-based risk
• Deploy fraud detection system as an API
• Monitor performance on live or real-world data

---

How to Run

1. Clone this repository
2. Install required packages:

   pip install -r requirements.txt