// Code will appear here

Connecting to Mode Selection

Establishing connection...

Launching...

Preparing environment...

Parametric Analysis of KNN

Systematically explore how different hyperparameters affect KNN classification performance. Configure K, sample size, distance metric, and scaling, then train the model to observe changes in metrics and the feature space visualization.
Click buttons to make a configuration and train model

K VALUE (Neighbours)

The number of nearest neighbours used for classification. Smaller K is more sensitive to noise (overfitting), larger K produces smoother boundaries (underfitting).

TRAINING SAMPLE SIZE

Number of training samples used. More samples generally improve classification but with diminishing returns. The test set remains fixed at 45 samples.

DISTANCE METRIC

Euclidean measures straight-line distance. Manhattan measures axis-aligned distance (sum of absolute differences). Click a test point after training to see the difference visually.

Euclidean Distance Manhattan Distance

FEATURE SCALING

Standard Scaling normalizes features to zero mean and unit variance. This prevents features with larger ranges from dominating the distance calculation. Observe the axis values change after training with scaling enabled.

No Scaling Standard Scaling

CURRENT CONFIGURATION

K Value6

Samples30

DistanceEuclidean

ScalingNone

Feature Space Plot

Train the model, then click a test point to see K nearest neighbours

Performance Metrics

Accuracy

Precision

Recall

F1 Score

Configure parameters and click TRAIN MODEL to see results

Analysis Scenarios

Effect of the Number of Neighbours (K)

Procedure:

Select different values of K (for example 6, 8 and 10).
Train the model for each value of K.
Observe the decision boundary visualization and performance metrics.
Record the values of Accuracy, Precision, Recall and F1-Score.

Expected Observations:

Smaller values of K produce irregular or noisy decision boundaries, leading to possible overfitting.
Larger values of K produce smoother boundaries, leading to possible underfitting.
Moderate values of K produce balanced boundaries, leading to better generalization.

Effect of Training Sample Size

Procedure:

Use the Sample Size buttons to change the number of training samples.
Train the model with different sample sizes (for example 30, 60, 90 and 105).
Observe how the performance metrics change.

Expected Observations:

When the number of training samples is small, the model has limited information, leading to unstable predictions and irregular decision boundaries.
As the number of training samples increases, the classifier captures the data distribution more effectively, improving classification performance.
Performance metrics such as Accuracy, Precision, Recall and F1-Score become more stable with more training data.
The decision boundary becomes more consistent and reliable as the model learns better class separation from a larger dataset.

Effect of Distance Metric on Classification

Procedure:

Select Euclidean Distance as the distance metric.
Train the model and record metrics.
Switch to Manhattan Distance while keeping the same value of K.
Train again and observe the changes in classification predictions and decision boundaries.

Expected Observations:

Different distance metrics may select different neighbouring points for the same test sample because the method of distance calculation varies.
Due to this difference in neighbour selection, the predicted class for some samples may change when the distance metric is altered.
Compare the performance metrics to determine which distance metric provides better classification performance for the given dataset.

Effect of Feature Scaling

Procedure:

Select No Scaling as the scaling option.
Train the KNN model and record metrics.
Switch to Standard Scaling.
Train again and observe the decision boundary and performance metrics.

Expected Observations:

Without feature scaling, features with larger numerical values may dominate the distance calculation, which can influence neighbour selection and reduce classification performance.
After applying feature scaling, all features contribute more equally to the distance computation, which may improve the classification results.
The decision boundary often becomes more balanced and consistent when normalization is applied.

Deep Dive Playground

Initializing core...

KNN Visualizer Playground

This simulation environment facilitates the empirical analysis of K-Nearest Neighbors classification dynamics. It visualizes the perturbation of performance metrics concerning hyperparameter modulation and progressive sample acquisition, whilst maintaining a static holdout set to isolate generalization behavior.

K VALUE

The number of nearest neighbours used for classification. Smaller K is more sensitive to noise (overfitting), larger K produces smoother boundaries (underfitting).

Number of Neighbors 6

TRAINING SAMPLES

Number of training samples used. More samples generally improve classification but with diminishing returns. The test set remains fixed at 45 samples.

Sample Count 0/105

DISTANCE METRIC

Euclidean measures straight-line distance. Manhattan measures axis-aligned distance (sum of absolute differences). Click a test point after training to see the difference visually.

Euclidean Manhattan

FEATURE SCALING

None Standard

Opens a grid view comparing all feature combinations (e.g., Sepal vs Petal) to understand class separability.

AUTO-PILOT

Automatically adds training samples and updates the model to visualize learning progression over time.

Animation Speed 1.3x

LIVE TESTING METRICS

Accuracy -

Precision -

Recall -

F1 Score -

ROC-AUC -

OVERVIEW

Dataset: Fisher's Iris

Features: 4 morphological measurements (cm)

Classes: 3 (setosa, versicolor, virginica)

Samples: 150 (balanced: 50 per class)

Current Split: 70-30 (105 / 45)

Increment: 5 samples

PCA Decision Boundary

Visualizes the decision regions generated by the KNN classifier projected onto the first two principal components. Click on a test point to see its K nearest neighbors.

Classes

Setosa

Versicolor

Virginica

Train

Test (Correct)

Test (Misclassified)

Testing Metrics Evolution

Metrics: accuracy, precision, recall, F1 score, and ROC-AUC tracked over sample growth. F1 score overlaps Accuracy closely.

ROC Curve (Test Set)

ROC (Receiver Operating Characteristic): Plots TPR vs. FPR. Higher AUC (Area Under Curve) means better performance.

STEPS

Select Distance Metric

Euclidean Distance

Manhattan Distance

Connecting to Mode Selection

KNN Experiment Modes