Bias-Variance Tradeoff Explorer (AKMproject)

Current project progress: Final Implementation Fully Deployed!

🔗 Live Demo: View our Shiny App on Posit Connect Cloud

---

Project Overview

This R package and interactive Shiny application provide a comprehensive toolkit for exploring the Bias-Variance Tradeoff through Monte Carlo simulations. It allows users to visualize how model complexity (Polynomial Degree or k-NN) affects error decomposition in real-time.

✨ Key Features (Final Milestone)

1. Interactive Error Decomposition

• Dynamic Hover Charts: Built with plotly, allowing users to see precise MSE, Bias², and Variance values by hovering over the curves.
• Model Selection: Compare Polynomial Regression and k-Nearest Neighbors (k-NN) models side-by-side.

2. Intelligent Model Recommendations

• Best Complexity Detection: The app automatically identifies the optimal complexity (e.g., "Degree 5") that minimizes Test MSE.
• Automated Feedback: Provides clear, text-based guidance on the recommended model and minimum test error achieved.

3. Advanced Monte Carlo Visualizations

• Prediction Spread: Visualizes model stability by plotting 100+ individual MC fitted curves against the true function.
• Smooth UX: Integrated shinycssloaders to provide visual feedback (spinners) during heavy simulation computations.

4. Data Transparency

• Exportable Results: Users can download the raw pointwise simulation data as a .csv for further independent analysis.

---

🛠️ Package Status & Compatibility

• Cross-Platform Tested: devtools::check() passed with 0 Errors / 0 Warnings on Windows 11, macOS, and Ubuntu 24.04 LTS.
• Dependency Management: Full environment reproducibility via manifest.json and a robust DESCRIPTION file.

In accordance with the course requirements, a detailed Generative AI Usage Statement is included in the package vignettes.

• Vignette Path: vignettes/ai_usage_statement.Rmd
• Final Update: Includes documentation on refactoring visualizations to plotly and implementing asynchronous UI feedback.