Machine Learning Pipeline Framework

A modular machine learning experimentation framework built in Python using scikit-learn.

The goal of this project is to provide a reusable architecture for training, evaluating, and comparing machine learning models across different datasets with minimal code changes.

Instead of creating one-off scripts for each dataset, the framework separates datasets, models, evaluation, and reporting into reusable components.

---

Features

• Object-oriented architecture
• Dataset abstraction
• Model abstraction
• Machine learning pipelines
• Classification and regression support
• Model comparison
• Feature importance / coefficient analysis
• Confusion matrix reporting
• Regression metrics (MAE, RMSE, R²)
• Factory pattern for model creation
• Factory pattern for evaluator creation
• Easily extensible for new datasets and models

---

Current Datasets

Titanic Survival Prediction

Classification problem.

Models:

• Logistic Regression
• Random Forest Classifier

Outputs:

• Accuracy
• Confusion Matrix
• Error Analysis
• Feature Analysis
• Model Agreement

---

California Housing Prices

Regression problem.

Models:

• Linear Regression
• Random Forest Regressor

Outputs:

• Mean Absolute Error (MAE)
• Root Mean Squared Error (RMSE)
• R² Score
• Feature Analysis
• Prediction Difference Analysis

---

Project Structure

machine-learning-pipeline/
│
├── data/
│
├── datasets/
│   ├── dataset.py
│   ├── titanic.py
│   └── california_housing.py
│
├── models/
│   ├── classification/
│   ├── regression/
│   ├── model.py
│   ├── model_factory.py
│   ├── model_type.py
│   └── problem_type.py
│
├── pipeline/
│   ├── machine_learning_pipeline.py
│   └── experiment_runner.py
│
├── evaluation/
│   ├── classification_evaluator.py
│   ├── regression_evaluator.py
│   └── evaluator_factory.py
│
├── results/
│
├── utils/
│
├── main.py
└── README.md

---

Design

The framework follows a modular pipeline:

Dataset
    │
    ▼
Preprocessing
    │
    ▼
Model Training
    │
    ▼
Prediction
    │
    ▼
Evaluation
    │
    ▼
Feature Analysis
    │
    ▼
Model Comparison

Core design principles include:

• Abstract base classes
• Factory pattern
• Separation of concerns
• Composition over inheritance where appropriate
• Extensible architecture

---

Technologies

• Python 3.12
• pandas
• NumPy
• scikit-learn

---

Future Improvements

• Wine Quality dataset
• Breast Cancer dataset
• Product Sales Forecasting
• K-Means clustering
• Cross-validation
• Hyperparameter tuning
• Model persistence
• Automatic report generation
• SHAP feature explanations

---

Example Output

MODEL COMPARISON

LINEAR_REGRESSION
MAE : 0.533
RMSE: 0.746
R²  : 0.576

RANDOM_FOREST_REGRESSOR
MAE : 0.328
RMSE: 0.506
R²  : 0.805

BEST MODEL
RandomForestRegressor

---

Author Notes

Developed as a personal project to explore machine learning engineering, reusable software architecture, and experiment management using Python.