This interactive application, built with Streamlit and Python, allows you to explore the main concepts of the k-Nearest Neighbors (kNN) algorithm in a visual and intuitive way.
kNN is one of the simplest Machine Learning methods, yet it can model non-linear decision boundaries.
This example focuses on kNN applied to classification tasks and was developed by Prof. Mariana Recamonde Mendoza as supporting material for the Machine Learning course taught at the Institute of Informatics — Federal University of Rio Grande do Sul (UFRGS).
🔗 https://inf-knn-app-explorer.streamlit.app
kNN has a very simple inductive bias:
Points with similar attribute values tend to be close in the feature space.
This explorer helps you visualize:
- How the choice of k changes the decision boundary
- The effect of different distance metrics
- How feature scaling influences the results
- The importance of feature normalization for kNN
- How the algorithm behaves with new, unseen data points
- The geometry of the data (moons, blobs), how it influences decisions, and how non-linear decision boundaries are formed
This tool is intended to serve as support for lectures, self-study, and live demonstrations.
The application has three main sections:
- Settings (sidebar)
- Decision boundary visualization
- Testing with new points (generalization)
In the sidebar, you can adjust:
- k (number of neighbors)
- Distance metric (Euclidean or Manhattan)
- Dataset (Moons or Blobs)
- Normalization (Min-Max scaling)
- Random seed for test points
- Generate new test points
These options allow you to create small experiments and immediately observe the effect of modeling decisions.
The app includes two main scenarios:
This scenario allows you to observe:
- How a small k produces irregular boundaries (tendency toward high variance)
- How a large k smooths the boundary (tendency toward high bias)
This scenario demonstrates that:
- Features with large values dominate the distance computation
- Normalizing features (for example, using Min-Max scaling) is essential in kNN
- Without normalization, the axis with the largest range “drives” the decision
The app automatically generates 10 new points (using their own random seed), which simulate unseen, new data.
You can:
- Select a point (1 to 10)
- Compare the true class and the predicted class for this point
- See the selected point highlighted in the plot
- View all numbered points in the feature space
This makes it clearer how the model behaves when making predictions on new data.
Author: Profa. Mariana Recamonde Mendoza.
📍 Institute of Informatics - Federal University of Rio Grande do Sul (UFRGS), Porto Alegre - RS, Brazil
The code was developed with the support of Generative AI (Gemini 3.1 and ChatGPT 5.2).