K-Nearest Neighbors — Interview Questions

A curated bank of conceptual, mathematical, and case-study interview questions on K-Nearest Neighbors.

How to use this page

Each question is tagged Easy / Medium / Hard. Try to answer before expanding the accordion. Questions are grouped by theme — Fundamentals, Distance & K, Curse of Dimensionality & Scaling, and Tricky cases.

1. Fundamentals

2. Distance Metrics & Choosing K

3. Curse of Dimensionality & Scaling

4. Tricky / Case-Study Questions

5. Advanced & Applied KNN

Quick-fire one-liners

Training time? O(1) — lazy learner, just stores data.
Default distance? Euclidean (L2).
Odd or even K? Odd, to avoid ties in binary classification.
Rule of thumb for K? K ≈ √n, then validate via CV.
Must do before fitting? Scale features.
Big-n speedup? KD-tree / Ball tree / FAISS / HNSW.
Why poor in high d? Distances concentrate — curse of dimensionality.

# Quick sklearn cheat-sheet for interviews
from sklearn.neighbors import KNeighborsClassifier, KNeighborsRegressor
from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import make_pipeline
from sklearn.model_selection import GridSearchCV

# Classifier with CV-tuned K and distance weighting
pipe = make_pipeline(StandardScaler(), KNeighborsClassifier())
grid = GridSearchCV(
    pipe,
    param_grid={
        "kneighborsclassifier__n_neighbors": [3, 5, 7, 11, 15, 21],
        "kneighborsclassifier__weights": ["uniform", "distance"],
        "kneighborsclassifier__metric": ["euclidean", "manhattan"],
    },
    cv=5, scoring="f1_macro", n_jobs=-1,
).fit(X_train, y_train)

# Regressor
reg = make_pipeline(StandardScaler(), KNeighborsRegressor(n_neighbors=10, weights="distance"))

# Approximate NN for large n
# from faiss import IndexFlatL2  # FAISS for million-scale lookup
# index = IndexFlatL2(d); index.add(X); D, I = index.search(query, k=10)

K-Nearest Neighbors — Interview Questions

1. Fundamentals

EasyQ1. What is K-Nearest Neighbors and how does it make predictions?

EasyQ2. Why is KNN called a 'lazy learner'?

MediumQ3. Is KNN parametric or non-parametric? What does that mean?

MediumQ4. What is the time and space complexity of KNN?

2. Distance Metrics & Choosing K

MediumQ1. What distance metrics can KNN use? When do you pick each?

MediumQ2. How do you choose K? What happens with very small or very large K?

MediumQ3. What is weighted KNN and when does it help?

EasyQ4. Why must you scale/standardize features before applying KNN?

3. Curse of Dimensionality & Scaling

HardQ1. What is the curse of dimensionality and how does it affect KNN?

HardQ2. How do KD-trees and Ball trees speed up KNN? When do they stop helping?

HardQ3. How do you make KNN scale to millions of points?

4. Tricky / Case-Study Questions

MediumQ1. How does KNN handle categorical features?

MediumQ2. Is KNN sensitive to imbalanced classes? How do you fix it?

EasyQ3. Can KNN be used for regression? How?

HardQ4. Why does KNN often perform poorly on text or very sparse high-dimensional data?

EasyQ5. What's the difference between KNN and K-Means?

MediumQ6. Your 1-NN model has 100% training accuracy but poor test accuracy. Why?

MediumQ7. How do you handle missing values in KNN?

EasyQ8. Pros and cons of KNN in one breath?

5. Advanced & Applied KNN

HardQ1. What is the Bayes error rate, and how does the 1-NN error relate to it?

MediumQ2. How does the bias-variance tradeoff manifest in KNN as K changes?

MediumQ3. How does KNN produce class probabilities? Are they well-calibrated?

EasyQ4. What happens at a tie vote in KNN classification? How do you break it?

MediumQ5. Can KNN handle online / streaming data?

HardQ6. What is Condensed Nearest Neighbor (CNN) and why use it?

HardQ7. How is KNN related to kernel density estimation (KDE)?

MediumQ8. How do irrelevant features hurt KNN, and how can you mitigate it?

HardQ9. Why is Mahalanobis distance sometimes preferred over Euclidean for KNN?

HardQ10. What is metric learning and how does it improve KNN?

MediumQ11. How does the choice of distance metric interact with feature scaling?

MediumQ12. Can KNN be parallelized? Where are the parallelism opportunities?

HardQ13. What is locality-sensitive hashing (LSH) and how does it help KNN?

MediumQ14. How would you build a recommender system using KNN?

MediumQ15. How would you detect anomalies with KNN?

Quick-fire one-liners