Machine Learning Examples

---

Table of Contents

• Setup
• Models
  • MNIST Digit Classification
  • Logistic Regression
  • Content-Based Filtering
  • Fraud Detection
    • Simple Model
    • Realistic Model
• License

Setup

1. Clone the repository:

git clone https://github.com/mineme0110/mnist.git
cd mnist

2. Create virtual environment:

python -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

Models

MNIST Digit Classification

Neural network for MNIST digit classification using PyTorch.

1. Install dependencies:

pip install -r requirements.txt

2. Run the training script:

python src/mnist/train.py

Logistic Regression

Binary classification example using scikit-learn's Logistic Regression.

1. Install dependencies:

pip install -r src/logistic-regression/requirements.txt

2. Run the training script:

python src/logistic-regression/train.py

This will demonstrate logistic regression on a generated dataset with visualization of the decision boundary.

Content-Based Filtering

Movie recommendation system using content-based filtering.

1. Install dependencies:

pip install -r src/content-based-filtering/requirements.txt

2. Run the recommendation system:

python src/content-based-filtering/train.py

This will demonstrate movie recommendations based on content similarity using a sample movie dataset. The system considers movie genres, actors, and descriptions to make recommendations.

Example output:

Getting recommendations for: The Dark Knight
Recommended Movies:
------------------------------------------------------------
1. Iron Man
   Genres: Action, Adventure, Sci-Fi
   Similarity Score: 0.8245
------------------------------------------------------------
2. The Matrix
   Genres: Action, Sci-Fi
   Similarity Score: 0.7856

Fraud Detection

XGBoost-based fraud detection system for identifying fraudulent transactions.

Prerequisites for Mac users:

# Install OpenMP library (required for XGBoost)
brew install libomp

Simple Model

A basic fraud detection model with clear separation between normal and fraudulent transactions.

1. Install dependencies:

pip install -r src/fraud-detection/requirements.txt

2. Run the simple fraud detection model:

python src/fraud-detection/train_simple.py

Features:

• Clear separation between classes
• Basic XGBoost parameters
• Perfect for understanding basic fraud detection concepts
• Shows idealized probability distributions

Realistic Model

A more sophisticated model that better represents real-world fraud detection scenarios.

1. Install dependencies (if not already installed):

pip install -r src/fraud-detection/requirements.txt

2. Run the realistic fraud detection model:

python src/fraud-detection/train_realistic.py

Features:

• Handles imbalanced classes
• Uses realistic transaction patterns:
  • Transaction amounts
  • Time of day patterns
  • Geographic distances
  • Transaction frequency
• Early stopping and validation
• Feature importance analysis
• More representative probability distributions

Example output:

Realistic Model Performance:
ROC AUC Score: 0.9985
Top 5 Most Important Features:
transaction_amount: 0.4532
time_of_day: 0.2876
distance_from_last: 0.1543
transaction_frequency: 0.0892
feature_5: 0.0157

License

This project is distributed under the terms of the MIT license.