In classification, which class does the likelihood principle choose?

The class with the highest likelihood (probability) for the object

The class with the most training records

The class whose centre is farthest away

What does the distance principle state?

Data that are close to each other are alike and probably in the same class or cluster

Data that are far apart are alike

Distance has no relation to class

Only categorical data can be compared

Why should attributes be normalised before computing distance?

So that likelihood becomes zero

To make all classes equal in size

To convert distance into probability

2065 Baisakh
Year not given · 3 marks

Short note: Likelihood and distance

Short answer

The likelihood-distance principle states that objects with high likelihood (class probability) or small distance are similar. Likelihood is used in Naive Bayes, distance in KNN, clustering, and outlier detection. Normalization is needed so one attribute does not dominate.

Write a short note on Likelihood and Distance (technique/principle).

Likelihood and Distance is a principle of data mining for finding patterns by asking how probable a pattern is and how close the data objects are. It is used by mining techniques to decide membership of an object in a group.

Likelihood

Likelihood is the probability (chance) that an object belongs to a class or model, given the data. In classification, we pick the class with the highest likelihood. Example: Naive Bayes compares P(X | class)·P(class) for each class. Techniques that follow the "most likely" outcome include Bayesian classification and statistical methods.

Distance

Distance is a numeric measure of how far apart two objects are in attribute space, e.g. Euclidean distance d(x, y) = √Σ(xᵢ − yᵢ)². The principle is: data that are close to each other are alike and probably belong to the same class or cluster.

Use

Classification: K-nearest neighbour gives a new object the class of its closest known objects.
Clustering: objects are grouped so that distance within a cluster is small and distance between clusters is large (k-means, hierarchical).
Outlier detection: objects far from all others, or with very low likelihood, are anomalies.

Attributes should be normalised so that a large-valued attribute does not dominate the distance.

✓ Answer checked by Claude Opus 5.5 (high effort)

DM techniques; likelihood and distance principle · Data Mining, chapter 3
Likelihood and distance; K-means drawbacks · Data Mining, chapter 3
"Do as your Neighbor Do" principle · Data Mining, chapter 3
Principle of optical distance measurement · Surveying II, chapter 2
Tacheometry principle, stadia distance and RL · Surveying II, chapter 2
Optical distance principle, tangential horizontal distance · Surveying II, chapter 2

All Data Mining chapter 3 questions →

Short note: Likelihood and distance

Likelihood

Distance

Use

Related questions