K Nearest Neigbour - Weighted KNN And Outlier Detection

Weighted KNN

Distance-weighted voting: Closer neighbors have more influence

from sklearn.neighbors import KNeighborsClassifier
 
knn = KNeighborsClassifier(
    n_neighbors=5,
    weights='distance', # Instead of 'uniform'
    metric='euclidean'
)

Benefits:

• Reduces impact of distant neighbors
• Better for variable density data
• Can use larger k values

KNN for Outlier Detection

KNN can detect anomalies based on distance to neighbors:

Method:

1. For each point, calculate distance to k-th nearest neighbor
2. Points with large k-distances are outliers
3. Threshold based on percentile or standard deviations

Applications:

• Anomaly Detection
• Novelty detection
• Quality control