Vehicle Silhouettes Machine Learning

Supervised and unsupervised machine learning models classifying vehicle silhouettes from geometric features, using Python (scikit-learn, Pandas, NumPy).

Tools & Skills Used

Quick Access

• Project Description
• Supervised Learning Notebook
• Unsupervised Learning Notebook
• Vehicle Silhouettes Dataset
• Cleaned Dataset

Setup & Installation

Prerequisites

• Python 3.12+
• uv package manager

1. Clone the Repository

git clone https://github.com/krauseannelize/ml-vehicle-silhouettes.git
cd ml-vehicle-silhouettes

2. Install Dependencies

uv sync

3. Run the Project

uv run jupyter lab

📌 Note: uv run automatically uses the project's virtual environment, no manual activation needed

Project Overview

Prospect Auto, a chain of car repair shops, requested models to classify vehicles based on silhouette features. This repo demonstrates two approaches:

• Supervised classification → Predicting vehicle classes using labeled data.
• Unsupervised clustering → Grouping vehicles without labels to discover natural structure.

Objectives

• Explore the Vehicle Silhouettes dataset.
• Build supervised models to classify vehicles.
• Apply unsupervised clustering methods to group vehicles.
• Compare results and assess which approach is most effective.

Methodology

1. Data Preparation

   • Load and clean dataset.
   • Normalize and standardize features.
   • Split into training/testing subsets.

2. Supervised Learning

   • Train classification models.
   • Evaluate with accuracy, precision, recall, F1‑score.

3. Unsupervised Learning

   • Apply PCA for dimensionality reduction.
   • Train clustering models (e.g., k‑means).
   • Evaluate with silhouette score and inertia.

Key Findings

• All supervised algorithms performed consistently well across classes, confirming strong generalization without significant overfitting.
• Logistic Regression emerged as the most balanced and reliable supervised model, achieving 95.1% accuracy on training data and 93.9% on test data.
• Unsupervised clustering (K-Means, Hierarchical, DBSCAN) revealed one dominant group with smaller overlapping sub-clusters that were imbalanced and did not align with true vehicle classes.
• While clustering is not a viable alternative for Prospect Auto's classification needs, it may still provide value for exploratory analysis or anomaly detection.