https://github.com/hase3b/knn-pipeline

This repository features a comprehensive ML pipeline for supervised learning tasks, covering data collection, cleaning, EDA, feature engineering, model training, and evaluation.
https://github.com/hase3b/knn-pipeline

classification cross-validation data-cleaning data-pipeline data-transformation eda feature-selection knn lazy-classifier lazy-predict model-evaluation pca regression supervised-learning

Last synced: 24 days ago
JSON representation

This repository features a comprehensive ML pipeline for supervised learning tasks, covering data collection, cleaning, EDA, feature engineering, model training, and evaluation.

• Host: GitHub
• URL: https://github.com/hase3b/knn-pipeline
• Owner: hase3b
• Created: 2024-02-19T21:35:33.000Z (11 months ago)
• Default Branch: main
• Last Pushed: 2024-07-15T23:32:37.000Z (6 months ago)
• Last Synced: 2024-07-16T03:06:27.399Z (6 months ago)
• Topics: classification, cross-validation, data-cleaning, data-pipeline, data-transformation, eda, feature-selection, knn, lazy-classifier, lazy-predict, model-evaluation, pca, regression, supervised-learning
• Language: Jupyter Notebook
• Homepage:
• Size: 3.12 MB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# KNN-Pipeline

This repository contains an implementation of a comprehensive machine-learning pipeline tailored for both classification and regression tasks. The pipeline encompasses various stages, including data collection, cleaning, transformation, exploratory data analysis (EDA), feature selection, dimensionality reduction, data splitting, cross-validation, model selection, training, and evaluation. Currently, the pipeline only supports the K-Nearest Neighbors (KNN) algorithm since the aim of the project is to demonstrate the pipeline execution and the pipeline can be easily modified to support other algorithms (at the end of pipeline V2 notebook an alternative approach to pipeline flow is demonstrated which allows for different algorithms without integrating them within the pipeline and even this approach is fairly simple and integration them within the pipeline flow would be fairly easily as well).

### Repository Structure

This repository contains Jupyter Notebooks for the pipeline's support and master functions along with the pipeline's execution on multiple datasets including both classification and regression problems. The notebook also contains detailed comments on each step of the pipeline as well as interpretation of EDA and results. Further, it also contains an excel file that shows the compiled results for pipeline V1.

## Pipeline Flow

1) Fetch Data: Data is sourced from the UCI Machine Learning Repository.

2) Data Cleaning: Treatment of missing and duplicate values to ensure data quality.

3) Exploratory Data Analysis (EDA): Numerical Summary: Summarizing numerical features to understand data distribution and characteristics.

4) EDA: Class Balance (%): Examining class distribution to identify potential imbalances (for classification tasks only).

5) EDA: Histogram & Box Plot of Important Numerical Features: Visualizing key numerical features to detect patterns and anomalies.

6) Outliers Detection & Treatment: Not Implemented Yet: Placeholder for future development.

7) Data Transformation: Encoding categorical variables and standardizing numerical features (both standard and min-max scaler available).

8) Feature Selection: Selecting the most relevant features using various methods.

   * Filter Methods: Includes correlation heatmap and mutual information.

   * Wrapper Methods:

       * Forward Selection: Iteratively adding features that improve model performance.

       * Backward Selection: Iteratively removing features that degrade model performance.

    * Dimensionality Reduction (PCA): Reducing the feature space while retaining most of the variance.

9) PCA Visualization: Visualizing the principal components to understand data separability (for classification tasks only).

10) Data Splitting & Cross-Validation: Splitting the data into training and testing sets and implementing cross-validation to ensure robust model evaluation.

11) Model Selection (Lazy-Predict): Utilizing lazy-predict approach to evaluate multiple models quickly and select the best performing one.

12) Model Training (KNN): Training the K-Nearest Neighbors (KNN) model.

    * Non-CV:

       * Classification

       * Regression

    * CV:

       * Classification

       * Regression

13) Model Evaluation: Evaluating model performance using various metrics.

    * Non-CV:

       * Classification: Precision, recall, F1-score.

       * Regression: Mean squared error (MSE), mean absolute error (MAE), root mean squared error (RMSE), R-squared, Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), p-value.

    * CV:

       * Classification: Evaluating classification performance with cross-validation.

       * Regression: Evaluating regression performance with cross-validation.

### Future Work

* Implement outliers detection and treatment.

* Expand the pipeline to include additional models and hyperparameter tuning.

* Enhance the visualization capabilities for better data insights.